DETAILED ACTION
Notice of Pre-AIA or AIA Status
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
2. This office action is responsive to Applicant’s Arguments/Remarks Made in an Amendment received on 06/08/2026.
Claim Rejections - 35 USC § 101
3. In view of Applicant’s argument [Remarks] and amendments filed 06/08/2026, claim rejection(s) with respect to 35 USC 101 have been fully considered and the rejection of claim 13 under 35 U.S.C. 101 is withdrawn.
Status of Claims
4. Claims 1-7, 9-19, 21 and 22 are pending in this application.
Claims 7, 9, and 11-14 are currently amended.
Claims 21 and 22 are newly added.
Response to Arguments
5. Regarding Applicant’s Argument (pages 7-10):
Applicant's arguments filed 08/26/2026 have been fully considered but are not
persuasive.
6. Regarding Applicant’s argument (pages 7-8):
Fang "only discloses reducing sample images by a window or masking function during training" and "is completely silent regarding generating partial images for input to an already-trained data-driven model".
This argument is not persuasive.
First, claim 1 does not recite a "trained" or "already-trained" data-driven model. Applicant's own Remarks (page 7) acknowledge this by bracketing the word "trained" when characterizing the claim: "providing the partial image to a [trained] data-driven model".
Second, claim 1 affirmatively ties the recited partial images to training. Claim 1 recites that "the data-driven model is parametrized according to a training data set including partial images and material information." The claim thus expressly contemplates that the partial images constitute the training data set. Fang's WxWxI windows, selected from the effective region and used to parametrize the deep neural network via cross entropy loss and small batch gradient descent (Col. 7), disclose such a training data set of partial images labeled with material information (Col. 3-4, 7). Applicant's observation that Fang's windowing occurs during training does not distinguish the claim, but establishes that Fang meets it.
Third, Fang expressly discloses a testing phase distinct from the training phase, and a
test set reserved for that phase. Fang: Col. 6: "the method according to embodiments of the
present disclosure may include a multi-modal data acquisition phase, a data pre-processing phase, a deep network training phase, and a testing phase."; Fang: Col. 4: "a set of data acquired by partly sampling the object that does not include data acquired at the middle posture as a test set...the test set may include 8 sets of data." Fang, Abstract and Col. 6, Block 103: the method generates "a material prediction result in testing by the material classification model." The material classification model that generates the prediction result in testing is, necessarily, an already-trained model.
Fourth, Fang's trained model is architecturally dimensioned to accept only a WxWxI partial image. Fang, Col. 7: "The size BxB of the convolution kernel of a first convolutional layer after the residual network may be adjusted based on a size of the input data. In a case where the size of the input data is WxW, the size of the convolution kernel may be set as W/16." A network whose convolution kernel is dimensioned as a function of W can receive only inputs of size WxW. Accordingly, the data supplied to Fang's already-trained material classification model during the testing phase is necessarily a WxWxI partial image produced by the same windowing operation described at Col. 7. Fang is therefore not silent on the point. Fang further states expressly that inference proceeds from windowed blocks: "it may be considered to learn based on the manner of the small window image block and to infer the material of the object using more spots" (Col. 9) (emphasis added). Fang thus contemplates inference, and not merely learning, from small window image blocks.
7. Regarding Applicant’s Argument (page 8):
Fang "considers a completely different paradigm," involving "capturing 17 raw images and 32 sets of data, performing computationally intensive pre-processing and normalizing," and that "[a] person of ordinary skill in the art would take none of the teachings of Fang as relevant when attempting to develop a minimally intensive process such as that of the present application".
This argument is not persuasive.
First, Fang expressly discloses the minimal, single-image configuration that Applicant
characterizes as absent from Fang. Fang, Col. 7: "The maximum of I may be 6, which may
uniquely represent a full-modal image obtained by combining the infrared modal, the depth
modal, normal angle modal and color modal. The minimum of I may be 1, which may represent
an infrared image or a grayscale image" (emphasis added). Fang further states: "A combination of all modal described above or a combination of a part of modal described above may be used as the network input." Fang therefore expressly discloses a partial image of dimension WxWx1
consisting of a single infrared speckle image as the input to the data-driven model. Applicant's
asserted distinction does not exist on the face of the reference.
Second, regarding the argument that Fang teaches away, a reference teaches away only when it "criticize[s], discredit[s], or otherwise discourage[s]" the claimed solution. In re Fulton, 391 F.3d 1195, 1201, 73 USPQ2d 1141, 1146 (Fed. Cir. 2004). Fang nowhere criticizes, discredits, or discourages the use of a single pattern image or a reduced processing pipeline. To the contrary, Fang states that its method is "simple and easy" (Col. 3, Col. 9) and expressly provides for single-modality input (Col. 7).
Third, that a reference discloses an embodiment more elaborate than the claimed
invention does not negate its teachings. A reference must be considered for all that it teaches to one of ordinary skill in the art, not merely for its preferred or fully instantiated embodiment. In re Heck, 699 F.2d 1331, 1332-33, 216 USPQ 1038, 1039 (Fed. Cir. 1983); MPEP 2123.
Fourth, claim 1 is drafted in open-ended form ("wherein the method comprises") and
does not exclude the acquisition of additional images, additional modalities, or additional
preprocessing. The recitation of "receiving the pattern image" does not preclude also receiving
other images. MPEP 2111.03. Fang's acquisition of additional modal images therefore does not
place Fang outside the scope of the claim.
Lastly, the advantage Applicant relies upon; "minimizing the computational burden of
material identification by capturing as few as a single image" (Remarks, p. 8), is not recited in
any pending claim. No pending claim recites capturing a single image, recites any limit on the
number of images received or recites any computational constraint.
8. Regarding Applicant’s Argument (page 9):
Applicant argues with respect to newly amended claims 11, 12 and new claim 22 that "Fang relies on manual input to identify regions of an image that depict an object vs. a background," that Fang "does not disclose any method of identifying an object within any of the color image, depth image, nor infrared image, such that background information is automatically suppressed," and that "it is unclear where in the process such a step could even be implemented with any reasonable expectation of success",
This has been considered but are moot because the arguments are addressed by the newly cited Collet Romea et al. reference explained in the body of rejection below. In particular Claims 11, 12, and new claim 22 are accordingly rejected at paragraph 19 above over Fang in view of Schick and further in view of Collet Romea, which supplies the automatic operation that Fang performs manually.
9. Regarding Applicant’s Argument (pages 9-10):
Applicant argues with respect to claim 14 that "Fang does not disclose nor render obvious" the recited "lower-dimensional representation," and that "Fang does not contemplate the motivation to reduce computational burdens during training or subsequent material identification".
This argument is not persuasive.
Fang expressly discloses the progressive generation of representations of successively lower dimensionality. Fang, Col. 7: the input "passes through four residual blocks that are gradually deepened in dimension but gradually reduced in size. Decrease of a size of an output is mainly caused by 2x2 mean value pooling at the end of the residual block. The pooling layer has the stride of 2 and the size of the output is thus reduced by half. After the output from the residual network passes through one convolutional layer, the size of output from the convolutional layer is limited to 1x1." Fang, FIG. 3, illustrates the same progression, terminating
in 1x1 convolutional layers of 2048 and 1024 channels and a classification output of dimension
C. The representation so produced is of substantially lower dimensionality than the WxWxI input partial image.
Second, the motivation to combine need not be found in the applied references, and need
not be the motivation that led Applicant to the claimed invention. KSR, 550 U.S. at 419-421;
MPEP 2144(IV). The motivation to reduce computational burden was expressly available in the art. Misra, of record, teaches extracting compact "patch representations" from image patches for the stated purpose of "reducing the amount of computational resources used" (Misra: Abstract; Description). Lastly, claim 14 recites its elements in the alternative ("and/or") and is met by Fang's teaching of partial images and of material information labelling the pattern features, irrespective of the amended second alternative.
Claim Rejections - 35 USC § 103
10. In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
11. The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
12. The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
13. The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
14. Claims 1-5, 7, 10, 13 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Fang et al. (US 11,782,285 B2, hereinafter "Fang") in view of Schick et al. (US 2022/0157044 A1, hereinafter "Schick").
Regarding Claim 1:
Fang discloses a computer implemented method for extracting material information of an object from a pattern image of the object (Fang: Abstract: "a material identification method and a device based on laser speckle and modal fusion"; Col. 1, lines 30-35: "The present disclosure relates to the fields of optical laser speckle imaging, computer vision, deep learning and material identification technologies"), wherein the method comprises:
receiving the pattern image showing the object while the object is illuminated with a light pattern (Fang teaches at Col. 3, Block 101: "data acquisition is performed on an object using a structured light camera for projecting laser speckles, to obtain a color modal image, a depth modal image and an infrared modal image"; the infrared modal image captures the laser speckle pattern reflected from the object surface.);
manipulating the pattern image to generate a partial image from the pattern image (Fang teaches at Col. 7: "A window having a size of W x W x I is randomly selected from an effective region satisfying the mask, where I is the number of channels of a combination of certain modal"; this windowed cropping of the acquired speckle image constitutes manipulating the pattern image to generate a partial image from the full pattern image.);
extracting material information of the object from the partial image by providing the partial image to a data-driven model (Fang teaches at Col. 7, Block 103: "inputting the color modal image, the depth modal image and the infrared modal image after the preprocessing into a preset depth neural network...to learn material characteristics from a speckle structure and a coupling relation between color modal and depth modal, to generate a material classification model for classifying materials, and to generate a material prediction result in testing by the material classification model"; the W x W partial images are the inputs provided to the deep neural network model. Fang further teaches at Col. 6 that the disclosed method comprises distinct training and inference phases: Fang teaches at Col. 4 that a dedicated test set is reserved for the testing phase: Fang further teaches at Col. 7 that the network is architecturally dimensioned for a WxW input: "The size BxB of the convolution kernel of a first convolutional layer after the residual network may be adjusted based on a size of the input data. In a case where the size of the input data is WxW, the size of the convolution kernel may be set as W/16." Because the trained material classification model is dimensioned as a function of W, the data provided to that model during the testing phase is necessarily a WxWxI partial image produced by the same windowing operation.),
wherein the data-driven model is parametrized according to a training data set including partial images and material information (Fang: Col. 7: "training the deep neural network may relate to data loading and the structure of the deep neural network"; Col. 7: the network is trained "by using a cross entropy as a loss function and by means of small batch gradient descent to obtain the material classification model based on the modal fusion"; Col. 3-4: training data includes multiple sets of data acquired for m objects formed of different materials, with W x W windows randomly selected as training samples labeled by material class. "The parameter C illustrated at the output layer may represent a total number of object types."); and providing the extracted material information (Fang: Abstract: "to generate a material prediction result in testing by the material classification model of the object"; Col. 9: the material classification model outputs material identification results.).
Fang teaches the use of a laser speckle pattern as the illumination pattern. However, Fang does not explicitly teach that the light pattern encompasses the broader range of illumination patterns such as point patterns, regular patterns, periodic patterns, and the like.
Schick teaches a detector and method for identifying at least one material property of an object (Shick: Abstract: "a detector for identifying at least one material property"), wherein an illumination source projects an illumination pattern comprising various pattern types including "at least one point pattern, in particular a pseudo-random point pattern, a random point pattern or a quasi random pattern, at least one Sobol pattern, at least one quasiperiodic pattern, at least one pattern comprising at least one pre-known feature, at least one regular pattern, at least one triangular pattern, at least one hexagonal pattern, at least one rectangular pattern" (para. [0038]). Schick further teaches that the illumination source may comprise "at least one laser source and one or more diffractive optical elements (DOEs)" for generating the illumination pattern (para. [0034]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the laser speckle-based material identification method of Fang to incorporate the various illumination pattern types as taught by Schick. One of ordinary skill in the art would have been motivated to make this modification because Schick teaches that using structured illumination patterns such as point patterns generated by laser and DOE configurations enables reliable identification of material properties from reflection beam profiles (Schick, para. [0004], [0034]), and both Fang and Schick are directed to the same field of endeavor of optically identifying material properties of objects using projected light patterns and analyzing the resulting reflection images. The substitution of one known illumination pattern type for another, or the use of additional pattern types, would yield the predictable result of providing a pattern image from which material information can be extracted. It is further noted that Schick is relied upon as cumulative evidence of the equivalence of known projected pattern types. Fang independently discloses the recited "light pattern" in the form of projected laser speckles comprising discrete spots (Col. 3, Block 101; Col. 8-9).
Regarding Claim 2:
The combination of Fang and Schick further teaches the method according to claim 1, wherein the pattern illumination is a regular pattern and/or a periodic pattern and/or a dot pattern and/or speckle pattern, with pattern features-and/or dots and/or speckles per face area at a distance of 5-200 cm.
Fang teaches that the pattern illumination is a speckle pattern, wherein data is acquired using "a structured light camera for projecting laser speckles" (Col. 3, Block 101). Schick further teaches that the pattern illumination may be "at least one point pattern" and/or "at least one regular pattern" and/or periodic and quasiperiodic patterns (para. [0038]), with pattern features projected onto an object at various distances.
Regarding the claimed distance of 5-200 cm, Fang expressly teaches an intended operating depth range that falls within the claimed range: "the 4 distances may include a distance range from 1 m to 1.5 m, a distance range from 1.5 m to 2 m, a distance range from 2 m to 2.5 m, and a distance range from 2.5 m to 3 m, covering most of intended depth range 0.8-3.5 m" (Col. 4) (emphasis added). Fang's disclosed intended depth range of 0.8-3.5 m includes 0.8 m (80 cm) through 2.0 m (200 cm), all of which lies within the claimed range of 5-200 cm. It is noted that where there is a finding of overlapping ranges, a prima facie case of obviousness exists. See In re Wertheim, 541 F.2d 257, 191 USPQ 90 (CCPA 1976); In re Woodruff, 919 F.2d 1575, 16 USPQ2d 1934 (Fed. Cir. 1990). The operating distance is further a result-effective variable that affects the resolution and density of pattern features on the object surface.)
It would have been obvious to one of ordinary skill in the art to optimize the operating distance of the system to fall within the claimed range of 5-200 cm, as the operating distance is a result-effective variable that affects the resolution and density of pattern features on the object surface and its optimization involves only routine skill. MPEP 2144.05(II). Furthermore, Schick teaches the detector may operate at various distances from the object.
Regarding Claim 3:
The combination of Fang and Schick teaches the method according to claim 1, wherein a pattern in the pattern image results from the object being illuminated with the pattern illumination, and/or wherein the pattern image comprises contributions from reflections of the pattern illumination and/or from a texture of the object.
Fang teaches that a pattern in the pattern image results from the object being illuminated with the pattern illumination and the pattern image comprises contributions from reflections of the pattern illumination and from a texture of the object (Col. 8-9: "the received speckle pattern is not in the form of a projected Gaussian bright spot and destructive interference and constructive interference exist in a central portion"; Col. 8: the material identification is based on "reflection and scattering characteristics of light on different object surfaces"; Col. 9: "a laser speckle image formed by interaction between spatial coherent light and microstructure on the surface of the material may encode material-level characteristics").
Schick further teaches that the reflection image includes "a feature in an image plane generated by the object in response to illumination" (para. [0030]) and that "human skin may have a reflection profile...comprising parts generated by back reflection of the surface...and parts generated by very diffuse reflection from light penetrating the skin" (para. [0041]).
Regarding Claim 4:
The combination of Fang and Schick teaches the method according to claim 1, wherein the partial image comprises at least parts of at least one pattern feature.
Fang teaches that the partial image comprises at least parts of at least one pattern feature, because the W x W window is "randomly selected from an effective region satisfying the mask" (Col. 7), where the effective region contains the laser speckle pattern features. The partial image inherently contains portions of the projected speckle pattern features since it is cropped from the speckle-illuminated image. Fang further teaches that "it may be considered to learn based on the manner of the small window image block and to infer the material of the object using more spots" (Col. 9), confirming that the windowed partial images contain speckle pattern features (spots).
Regarding Claim 5:
The combination of Fang and Schick teaches the method according to claim 1, wherein the material information is derived from reflections of the pattern illumination.
Fang teaches that the material information is derived from reflections of the pattern illumination (Col. 8: "a feasible way of using the optical signal for material identification is to use the reflection and scattering characteristics of light on different object surfaces to distinguish objects"; Col. 8: "Since different objects may have different subsurface reflection characteristic as well as different scattering and reflection characteristics, responses of different objects to incident light are different in aspects of time, space, and exiting angle"; Col. 8-9: the BRDF model describes the relation between incident and reflected light, and "Information about the material of the object...is included in the function fr(x,i,o,lambda,t)"). The deep neural network learns material characteristics from the speckle structure formed by the interaction between the projected laser light and the object surface, i.e., from the reflections of the pattern illumination.
Regarding Claim 7:
The combination of Fang and Schick teaches the method according to claim 1, wherein the step of extracting material information of the object from the partial image further comprises:
determining a lower-dimensional representation of the partial image (Fang teaches determining a lower-dimensional representation of the WxWxI partial image at Col. 7: "The input is subjected to feature extraction via a 3x3 convolutional layer, and passes through four residual blocks that are gradually deepened in dimension but gradually reduced in size. Decrease of a size of an output is mainly caused by 2x2 mean value pooling at the end of the residual block. The pooling layer has the stride of 2 and the size of the output is thus reduced by half. After the output from the residual network passes through one convolutional layer, the size of output from the convolutional layer is limited to 1x1. After the output from the convolutional layer passes through two convolutional layer having the convolution kernel with a size of 1x1, instead of a full-connection layer, a classification output having a dimension that equals to the number of classified materials is outputted" (emphasis added). Fang, FIG. 3, illustrates the corresponding progression: input x -> 3x3 conv, 64 -> residual block N -> residual block 64 -> residual block 128 -> residual block 256 -> residual block 512 -> 2x2 mean value pooling -> BxB conv, 2048 -> 1x1 conv, 1024 -> 1x1 conv, C. Fang thus expressly discloses stepwise reduction of the spatial dimension of the representation by half at each residual block, culminating in a 1x1 spatial output and a final representation of dimension C, which is of substantially lower dimensionality than the WxWxI input partial image),
wherein the representation is associated with a physical signature embedded in the reflections of the pattern illumination in the partial image (Fang further teaches that the representation is associated with a physical signature embedded in the reflections of the pattern illumination (Col. 8-9: "Theoretically, the interference information of material coding may be obtained according to the speckle pattern of a spot"; "The observed speckle pattern is formed by coherent superposition of subsurface scattering (volume scattering) and surface scattering of the projected light on the surface of the object"). The network's learned lower-dimensional representation necessarily captures these physical signatures, i.e., the BRDF and scattering characteristics, because the network must discriminate among them to output a correct material classification); and
providing the partial image to a data-driven model, wherein the data-driven model is parametrized according to the training data set further comprising the representation of one or more of the partial images and/or the cluster of one or more of the partial images, and wherein the data-driven model is parameterized to extract the material information (Fang teaches that the step of extracting material information further comprises providing the partial image to a data-driven model, wherein the data-driven model is parametrized according to the training data set further comprising the representation of one or more of the partial images (Col. 7: the deep neural network processes W x W input patches through "feature extraction via a 3 x 3 convolutional layer, and passes through four residual blocks that are gradually deepened in dimension but gradually reduced in size"; the intermediate feature maps generated by the convolutional layers and residual blocks constitute learned representations of the partial images; the training dataset comprises multiple W x W partial images with associated material labels), and wherein the data-driven model is parameterized to extract the material information (Col. 7: the network outputs "a classification output having a dimension that equals to the number of classified materials"; the model is parameterized via cross-entropy loss and gradient descent to extract material class information).
The same motivation to combine Fang with Schick as set forth in the rejection of claim 1
applies equally here.
Regarding Claim 10:
The proposed rejection of Claim 1 over Fang and Schick is similarly applied to reject
claim 10. Fang teaches the corresponding apparatus structure, including acquisition module 100 (receiving unit), preprocessing module 200 (manipulating unit), and neural network prediction module 300 (extracting unit and provisioning unit) (FIG. 6, Col. 9-10), as well as an electronic device comprising a processor and memory (FIG. 7, Col. 10).
Regarding Claim 13:
The combination of Fang in view of Schick teaches at least one non-transitory computer-
readable storage medium storing thereon a computer program for extracting material information of an object from a pattern image of the object, the computer program including computer-executable instructions that cause a computer to execute the steps of the method according to claim 1 (Fang, Col. 10: "Embodiments of the present disclosure provide a non-transitory computer readable storage medium, having one or more computer programs stored thereon. When the one or more computer programs are executed by a processor, the method described above is executed"; FIG. 7: electronic device with processor and memory storing computer program instructions).
Regarding Claim 14:
The combination of Fang and Schick teaches a data structure product comprising manipulated image data for extracting material information from a pattern image (Fang, Col. 3-7: the training dataset comprises preprocessed, windowed image data used for material classification. Col. 10 and FIG. 7: the data is held in system memory 28 of the electronic device during training and inference),
wherein the manipulated image data includes partial images (Fang, Col. 7: W x W windowed image patches randomly selected from the effective region of the preprocessed multi-modal images; these partial images constitute manipulated image data.);
and/or at least one lower-dimensional representation of the partial image, the representation associated with a physical signature of the partial image; (Fang, Col. 7 and FIG. 3: the convolutional neural network generates representations of the partial images that are reduced in spatial dimension by half at each residual block and are ultimately "limited to 1x1," as set forth in the rejection of claim 7 above; those representations are associated with physical signatures encoded in the speckle pattern reflections (Col. 8-9));
and/or at least one cluster associated with the representation and/or material information labelling the pattern features in the pattern image (Fang, Col. 3-4, 7: each partial image in the training set is labeled with a material class; Col. 7: "the parameter C illustrated at the output layer may represent a total number of object types"; training data associates material class labels with the pattern image data). Note: Claim 14 recites these elements in the alternative ("and/or"), and at least the first and last elements are taught by Fang. Accordingly, the claim is met by prior art teaching any one of the recited alternatives. Fang teaches at least the first alternative (partial images), the second alternative (lower-dimensional representation), and the fourth alternative (material information labelling). The amendment to the second alternative therefore does not place the claim in condition for allowance).
15. Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Fang in view of Schick, and further in view of Shao et al. (CN 106407914 A, hereinafter "Shao").
Regarding Claim 6:
The combination of Fang in view of Schick discloses the method according to claim 1, wherein the object is (Fang teaches that the material information comprises a material class (Col. 7: "a classification output having a dimension that equals to the number of classified materials"; the system outputs a material class prediction for the object).
Fang in view of Schick do not explicitly teach that the object is a face.
Schick teaches that the detector may be configured for "detection of biological tissue, in particular human skin" (para. [0015]), and that material classifiers include "skin or non-skin" and "biological or non-biological material" (para. [0014]).
Shao teaches that the object is a face (Shao teaches a method for detecting a face using an infrared structured light emitter that projects infrared light onto the face to form a speckle pattern (Step S230: "obtaining the face under the irradiation of infrared light formed by the speckle pattern"), wherein texture information (i.e., material property of the face surface) is obtained from the speckle pattern (Step S240: "different material structure in the structured light can form different speckle pattern. [The] processor can obtain the texture information of the face according to the received speckle pattern, namely the material property of the face surface"), and the material information is used to determine whether the face belongs to a living body (Step S250: "if the texture information in accordance with human skin texture distribution rule and the depth information meets the face depth distribution rule, it is determined that the face belonging to the living body"). Shao thus explicitly teaches the object is a face, and the material information comprises a material class (skin vs. non-skin) and a material property (texture distribution).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have applied the material identification method of Fang, as modified by Schick, to the face detection and liveness verification application taught by Shao. One of ordinary skill in the art would have been motivated to make this modification because Shao teaches that extracting material/texture information from infrared speckle patterns projected onto a face enables reliable living body detection for biometric security applications (Shao, pp. 7-8), and applying Fang's deep learning-based material classification model to face material analysis would improve the accuracy and robustness of face liveness detection compared to rule-based texture matching, as deep learning models can learn more discriminative material features from speckle patterns. See KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398, 415-421 (2007) (applying a known technique to a known device ready for improvement to yield predictable results is indicative of obviousness).
16. Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Fang in view of Schick, and further in view of Misra et al. (US 10,013,637 B2, hereinafter "Misra").
Regarding Claim 9:
The combination of Fang and Schick teaches the method according to claim 7, wherein the partial image is part of a data set of partial images; and
Fang teaches that the partial image is part of a data set of partial images (Col. 3-4, 7: multiple W x W windows are selected across multiple objects of different materials to form training, verification, and test sets; each material class has multiple sets of data forming a dataset of partial images). Fang further teaches determining a representation of the partial image, wherein the representation is associated with a physical signature embedded in the reflections of the pattern illumination in the partial image (Col. 7-9: the convolutional layers and residual blocks extract feature representations that capture physical signatures, i.e., the BRDF/scattering characteristics, encoded in the speckle pattern reflections, as discussed in the rejection of claim 8).
Fang in view of Schick do not teach that a cluster is derived via clustering of the partial images of the data set based on the representation, wherein the cluster is associated with the representation.
Misra teaches that a cluster is derived via clustering of the partial images of the data set based on the representation, wherein the cluster is associated with the representation.
Misra teaches a method of optimizing multi-class image classification using patch features (Abstract: "Optimizing multi-class image classification by leveraging patch-based features extracted from weakly supervised images to train classifiers is described"), wherein the method comprises:
extracting one or more patches from individual images (Claim 1: "extracting one or more patches from individual weakly supervised images of the plurality of weakly supervised images"; the patches constitute partial images extracted from full images, analogous to Fang's W x W windows);
extracting patch-based feature representations from the patches (Claim 1: "extracting patch-based features from the one or more patches; extracting patch representations from individual patches of the one or more patches"; the patch-based features and patch representations are learned representations of the partial images);
deriving clusters via clustering of the patches based on the representations (Claim 1: "arranging individual patches into a plurality of clusters based at least in part on the patch-based features"; the patches are clustered based on their extracted feature representations), wherein the clusters are associated to the representations (Claim 1: the individual clusters correspond to groupings of patches with similar feature representations; the clusters are used to "remov[e] at least some of the individual patches from at least one cluster of the plurality of clusters" based on similarity values, demonstrating that clusters are defined by and associated with the underlying patch representations); and
using the clustered data to train classifiers (Description: the techniques leverage patch-based features to optimize the multi-class image classification by improving accuracy in using classifiers to classify incoming images and reducing the amount of computational resources used).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated the patch-based clustering technique of Misra into the material classification training pipeline of Fang, as modified by Schick. One of ordinary skill in the art would have been motivated to make this modification because Misra teaches that clustering image patches based on their extracted feature representations and using the resulting cluster structure to refine the training data improves classification accuracy while reducing computational resources (Misra, Abstract; Description). Applying Misra's clustering of patch representations to Fang's W x W partial images and their learned feature representations would predictably improve the organization of Fang's material classification training data by grouping partial images with similar physical signatures, enabling better identification of material sub-categories and removal of outlier training samples that could degrade model performance. Both Fang and Misra are directed to image classification using patch/window-based feature extraction and deep learning, and combining their teachings involves applying a known technique (patch clustering for classification optimization) to a known method (speckle-based material classification using partial images) to yield the predictable result of improved training data organization and classification accuracy. See KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398, 415-421 (2007).
17. Claims 15-16, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Fang in view of Schick, and further in view of Shao.
Regarding Claim 15:
The combination of Fang and Schick teaches the method of using the manipulated data according to claim 14 but do not expressly disclose the method comprising using the manipulated data to authenticate an object.
Shao teaches the method comprising using the manipulated data to authenticate an object (Shao teaches a method of using manipulated data to authenticate an object, specifically using depth information and texture/material information obtained from an infrared speckle pattern to perform living body detection on a face (Step S250: "combining the depth information and the texture information determining whether the face belongs to living body"), followed by face recognition wherein the face is compared to "the identity card face" to "determine whether the face [is] consistent with the card face" (p. 10), or compared to "known face in the first database, to determine whether the face is one of known face in the first database" (p. 10). The combination of liveness detection and face recognition constitutes authentication of the object (face)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have used the manipulated image data (partial images with material labels) of the combination of Fang and Schick for the purpose of authenticating an object as taught by Shao. One of ordinary skill in the art would have been motivated to make this modification because Shao teaches that material/texture information derived from speckle patterns is effective for distinguishing real faces from spoofing attacks such as masks, photos, and videos (Shao, p. 2: "based on the detected mode of general camera user needs to do some action at random, difficulty is high, time is long, and the attack of the mask, photo, video and paper is not easily identified"), and using Fang's deep learning-based material classification data to support face authentication would provide a more robust and accurate authentication system.
Regarding Claim 16:
A method of using the manipulated data obtained by the method according to claim 1, the method comprising using the manipulated data to authenticate an object.
this claim recites a method of using the manipulated data obtained by the method according to claim 1 to authenticate an object.
The combination of Fang, Schick, and Shao teaches this limitation for substantially the same reasons as set forth in the rejection of claim 15. The manipulated data obtained by the method of claim 1 (i.e., partial images processed through the data-driven model to extract material information) is used for authentication as taught by Shao's living body detection and face recognition method. The same motivation to combine applies.
Regarding Claim 19:
A method of using the manipulated data obtained by the method according to claim 1, the method comprising using the manipulated data to biometrically authenticate an object.
This claim recites a method of using the manipulated data obtained by the method according to claim 1 to biometrically authenticate an object. The combination of Fang, Schick, and Shao teaches this limitation. Shao explicitly teaches biometric authentication of a face using speckle-derived texture/material information and depth information: the system performs living body detection (Step S250) followed by face recognition comparing the face to identity card photos or a database of known faces (pp. 9-10). This constitutes biometric authentication.
It would have been obvious to apply the material information extracted by the method of Fang (as modified by Schick) to the biometric face authentication system of Shao for the same reasons discussed in the rejection of claim 15.
18. Claims 17 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Fang in view of Schick, and further in view of Zheng (US 2019/0364226 A1, hereinafter "Zheng").
Regarding Claim 17:
The combination of Fang and Schick teaches the method according to claim 1, but do not expressly disclose wherein the pattern illumination is a regular pattern and/or a periodic pattern and/or a dot pattern and/or speckle pattern with at least 2 pattern features and/or dots and/or speckles per face area at a distance of 30-60 cm.
Zheng discloses wherein the pattern illumination is a regular pattern and/or a periodic pattern and/or a dot pattern and/or speckle pattern with at least 2 pattern features and/or dots and/or speckles per face area at a distance of 30-60 cm.
Zheng teaches a dot projector for face authentication applications (Abstract: "a dot projector comprising a movable base, a light source emitter disposed above the movable base, a collimator, located in a front side of the light source emitter, a diffractive optical element (DOE) located in the front side of the light source emitter"), wherein:
the dot projector emits a plurality of light spots projected onto a human face (Description: "After the light source generated by the light emitter passes through the diffractive optical element, multiple light spots will be generated to the objects or to the face"; Description: the dot projector projects more than thousands or tens of thousands of light spots onto the face);
the number of projected dots per face is in the thousands (Claim 5: "the number of light spots is less than 10,000"; Description: the system projects "more than thousands" of light spots onto the face surface); and
the dot projector operates at face-to-device distances typical of mobile phone use, i.e., approximately 20-50 cm (Description: the dot projector is designed for integration into mobile phones for face recognition; the infrared camera and dot projector are configured for typical phone-to-face operating distances; Description: "electronic products with face recognition functions have been continuously developed...mobile phones").
Since a typical human face has an area of approximately 400-600 cm2, Zheng's projection of thousands of light spots onto a face at phone-to-face distances (approximately 20-50 cm, which falls within the claimed 30-60 cm range) results in a dot density far exceeding 2 pattern features per face area. For example, even conservatively estimating 2,000 dots across a 500 cm2 face yields approximately 4 dots per cm2, well in excess of 2 features per face area.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have configured the structured light illumination system of Fang, as modified by Schick, to project a dot pattern at the density and distance taught by Zheng. One of ordinary skill in the art would have been motivated to make this modification because Zheng teaches that projecting thousands of dots onto a face at short range via a laser and DOE configuration enables accurate 3D face contour calculation for face recognition (Zheng, Description), and configuring Fang's laser speckle projector to operate at such densities and distances would enable the material identification system to be applied to face-directed applications such as face authentication and liveness detection, which is a desirable and commercially important application of material identification technology. Furthermore, the specific density and distance parameters are result-effective variables that a person of ordinary skill in the art would routinely optimize based on the desired spatial resolution of material classification and the intended use case. See In re Aller, 220 F.2d 454, 456, 105 USPQ 233, 235 (CCPA 1955); MPEP 2144.05(II).
Regarding Claim 18:
The proposed combination of Fang in view of Schick discloses the method according to claim 1, wherein the pattern illumination is a regular pattern and/or a periodic pattern and/or a dot pattern and/or speckle pattern
Fang in view of Schick do not expressly disclose at least 6 pattern features and/or dots and/or speckles per face area at a distance of 15-80 cm.
Zheng discloses at least 6 pattern features and/or dots and/or speckles per face area at a distance of 15-80 cm.
The combination teaches that the pattern illumination is a speckle pattern (Fang), may include regular, periodic, and dot patterns (Schick), and that a dot projector for face applications projects thousands of dots onto a face at short range (Zheng).
Claim 18 requires at least 6 pattern features and/or dots and/or speckles per face area at a distance of 15-80 cm. Zheng teaches the projection of "more than thousands or tens of thousands" of light spots onto a face (Description), with Claim 5 reciting "the number of light spots is less than 10,000." At the broader distance range of 15-80 cm, the phone-to-face operating distance of Zheng's dot projector falls squarely within this range. Using the same calculation as above, thousands of dots across a face area of approximately 400-600 cm2 yields a density far exceeding 6 features per face area. Even at the outer distance of 80 cm, where the projected pattern would spread over a larger area but still substantially illuminate the face, the density of thousands of dots would satisfy the requirement of at least 6 pattern features per face area.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to configure the system to provide at least 6 pattern features per face area at a distance of 15-80 cm, for the same reasons as discussed in the rejection of claim 17. One of ordinary skill in the art would have recognized that increasing pattern density improves the spatial resolution and accuracy of material classification, as more pattern features per unit area provide more data points for the data-driven model to analyze material characteristics. The claimed density of at least 6 features per face area represents a modest threshold well within the capabilities of commercial structured light systems as evidenced by Zheng's teaching of thousands of projected light spots. Furthermore, the distance range of 15-80 cm encompasses standard operating distances for mobile face recognition systems as taught by Zheng, and optimizing the system to operate within this range involves routine engineering considerations. See In re Aller, 220 F.2d 454, 456, 105 USPQ 233, 235 (CCPA 1955).
19. Claims 11, 12, and 22 are rejected under 35 U.S.C. 103 as being unpatentable over Fang
in view of Schick, and further in view of Collet Romea et al. (US 2019/0379873 A1, hereinafter
"Collet Romea").
Regarding Claim 11:
The combination of Fang and Schick teaches a computer implemented method for
training a data-driven model suitable for extracting material information of an object from a pattern image of the object, wherein the method comprises receiving the pattern image showing the object while the object is illuminated with a light pattern; manipulating the pattern image to generate a partial image; training a data-driven model to determine material information of the object from the partial image, wherein the training comprises parametrizing the data-driven model according to a training data set including partial images and material information; and providing the trained data-driven model (Fang, Col. 3, Block 101 (receiving); Col. 7 (windowing to generate the partial image); Col. 6-7, Block 103 (training the preset depth neural network to learn material characteristics from the speckle structure); Col. 7 (parametrizing via cross entropy loss and small batch gradient descent on labeled WxWxI windows); Abstract and Col. 10, FIG. 7 (providing the trained material classification model, stored on an electronic device comprising a processor and memory), as set forth in the rejection of claim 1 above.
Claim 11 has been amended to further recite that the manipulating comprises
"automatically suppressing and/or removing background information included in the pattern image."
Fang teaches suppressing and removing background information included in the pattern
image, but teaches doing so by manual means rather than automatically (Fang, Col. 6: "The infrared modal image, the depth modal image, the normal angle modal image, and the color modal image (three channels) are stacked to form a sample data map of a size of 1280x960 pixels. In addition, a seventh binarization mask channel is manually marked to indicate a region of the object in the image, and the background is excluded". Fang, Col. 7: the WxWxI window is thereafter "randomly selected from an effective region satisfying the mask," i.e., window selection is confined to the object region to the exclusion of the excluded background.
Fang in view of Schick do not expressly teach that the suppression and/or removal of the background information is performed automatically.
Collet Romea teaches automatically suppressing and removing background information
included in an image of a scene illuminated with a projected light pattern.
(Collet Romea, para. [0025]: "FIG. 1 shows an example system in which a pod 100 comprising stereo IR cameras 101 and 102, stereo RGB cameras 103 and 104, and a projector 106 (e.g., an IR laser diffracted into many thousands of dots) captures one or more frames of stereo (e.g., clean) IR images 108, RGB images 109 and depth data 110 (e.g., stereo images of the projected light pattern)." The pattern image of Collet Romea is thus the same class of image as the infrared modal image of Fang. Collet Romea, para. [0028]: "a projector 106 is shown that projects an IR pattern onto a scene, such as a pattern of spots (e.g., dots) or a line pattern, although other spot shapes and/or pattern types may be used. ... By illuminating the scene with a relatively large number of distributed infrared dots, the IR cameras 102 and 103 capture texture data as part of the infrared depth image data 110." Collet Romea, para. [0030]: "The image processing system or subsystem 120 includes a processor 121 and a memory 122 containing one or more image processing algorithms, including a multimodal, multi-cue foreground background segmentation algorithm 124 as described herein. In general, the segmentation algorithm 124 outputs a set of per-pixel probability data 126, representative of whether each pixel is likely to be a foreground or background pixel. The pixel probability data 126 is input into a global binary segmentation algorithm 128 (e.g., a Graph Cuts algorithm), which uses the pixel probability data 126 as a data term to segment the image into a segmented image 130, e.g., the foreground only as part of a stream of segmented images". This is an automatic determination of background information within the pattern image, followed by automatic removal of the background such that only the foreground object remains. Collet Romea, para. [0048]: "background subtraction 334 is performed on the before and after IR images to obtain the contribution factor D3 for each pixel." See also FIG. 3; FIG. 7, step 716 ("Get D3 via Background IR Subtraction"). Collet Romea, para. [0071]: "face / person detection may be used as another factor." Collet Romea expressly frames this automation as an improvement over prior manual and setup-dependent techniques (paras. [0003-0005]: chroma-keying "generally needs very controlled conditions" and "is limited to situations where a screen can be placed in the background, which is often not practical or possible"; background subtraction "has problems in disambiguating areas in which the foreground and background are similar"; depth-based segmentation "is not sufficient in many scenarios"), and states that "the framework is configured to exploit different modalities of information to achieve more robust and accurate foreground / background segmentation results relative to existing solutions" (para. [0021])).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the training method of Fang, as modified by Schick, such that the mask that excludes the background is generated automatically by the foreground/background segmentation of Collet Romea, thereby automatically suppressing
and/or removing background information included in the pattern image. One of ordinary skill in the art would have been motivated to make this modification for the following reasons. First, Collet Romea itself identifies the problem being solved, existing single-mode techniques "are not particularly robust" (para. [0005]), and teaches that automatic multimodal segmentation yields "more robust and accurate foreground / background segmentation results relative to existing solutions" (para. [0021]). Second, Fang already requires a mask that excludes the background (Col. 6) and confines window selection to the resulting effective region (Col. 7); automating the generation of that mask eliminates the per-image manual marking that Fang requires and thereby permits acquisition of substantially larger training sets, which directly benefits Fang's deep neural network. Third, background pixels carry no material information about the object under test, and excluding them prevents contamination of the training data and consequent degradation of classification accuracy. Fourth, Collet Romea operates on the same class of image data as Fang and Schick, namely an infrared image of a scene illuminated with a projected IR-dot pattern (Collet Romea, paras. [0025], [0028]; Fang, Col. 3, Block 101; Schick, para. [0038]), such that its segmentation technique is directly applicable to Fang's acquired images with a reasonable expectation of success. Fang's manually marked mask channel accomplishes the identical result recited in the claim, namely that "the background is excluded" (Fang, Col. 6). Automating the generation of that mask to accomplish that same result would have been obvious to one of ordinary skill in the art.
Regarding Claim 12:
Claim 12 recites an apparatus corresponding to the method of claim 11, comprising a
receiving unit, a manipulating unit configured to perform the manipulating "comprising
automatically suppressing and/or removing background information included in the pattern
image," a training unit, and a provisioning unit.
Fang teaches the corresponding apparatus structure, including acquisition module 100 (receiving unit), preprocessing module 200 (manipulating unit), and neural network prediction module 300 (training unit and provisioning unit) (FIG. 6, Col. 9-10), as well as an electronic device comprising a processor and memory (FIG. 7, Col. 10). The manipulating unit corresponds to preprocessing module 200, which performs the masking and windowing operations of Col. 6-7.
Collet Romea teaches the corresponding structure for automatic background suppression: "The image processing system or subsystem 120 includes a processor 121 and a memory 122 containing one or more image processing algorithms, including a multimodal, multi-cue foreground background segmentation algorithm 124" (para. [0030]; FIG. 1).
The automatic background suppression limitation is taught by Collet Romea and
rendered obvious for the reasons set forth in the rejection of claim 11 above.
Regarding Claim 22:
Claim 22 recites the method according to claim 1, wherein manipulating the pattern image to generate the partial image comprises automatically determining background information within the pattern image; and generating the partial image by suppressing the background information and/or cropping the pattern image to remove portions of the pattern image including the background information.
automatically determining background information within the pattern image
(Collet Romea, FIG. 7, steps 702-716, teaches an explicit sequential process of automatically determining background information within the pattern image: capture background IR (step 702) -> capture current IR (step 706) -> select a pixel (step 709) -> "Get D3 via Background IR Subtraction" (step 716). Collet Romea, para. [0048]: "background subtraction 334 is performed on the before and after IR images to obtain the contribution factor D3 for each pixel." Collet Romea, para. [0030]: the segmentation algorithm "outputs a set of per-pixel probability data 126, representative of whether each pixel is likely to be a foreground or background pixel.")
generating the partial image by suppressing the background information and/or cropping the pattern image to remove portions of the pattern image including the background information:
(Collet Romea, para. [0030]: the per-pixel probability data is used "as a data term to segment the image into a segmented image 130, e.g., the foreground only." Fang teaches cropping the pattern image to a sub-region that excludes background portions: the WxWxI window is selected "from an effective region satisfying the mask" (Col. 7), where the mask is the channel by which "the background is excluded" (Col. 6)).
Claim 22 recites these operations in the alternative ("and/or"), such that the claim is met by prior art teaching either suppression or cropping. The combination teaches both. The same motivation to combine Fang, Schick, and Collet Romea as set forth in the rejection of claim 11 applies equally here.
20. Claim 21 is rejected under 35 U.S.C. 103 as being unpatentable over Fang in view of Schick.
Regarding Claim 21:
Claim 21 recites the method according to claim 1, wherein manipulating the pattern image to generate the partial image comprises determining a maximum of an intensity distribution [of] the pattern image; and cropping the pattern image to include the maximum of the intensity distribution as part of the partial image.
The combination of Fang and Schick teaches all limitations of claim 1 as set forth at
paragraph 14 above. Fang teaches that the projected pattern comprises discrete spots, that each spot is a localized region of elevated intensity relative to the surrounding unilluminated image area, and that the material-encoding information resides at and within those spots. Fang, Col. 8-9: "As illustrated in FIG. 4, as can be seen from an enlarged high-definition speckle pattern, the received speckle pattern is not in the form of a projected Gaussian bright spot and destructive interference and constructive interference exist in a central portion. Material classification of the object may be achieved with this information. Theoretically, the interference information of material coding may be obtained according to the speckle pattern of a spot. However, since the acquired speckle pattern has a low resolution, it may be considered to learn based on the manner of the small window image block and to infer the material of the object using more spots". Fang, Col. 7: the WxWxI window is "randomly selected from an effective region satisfying the mask," i.e., from the region of the image containing the projected spots. Fang thus teaches cropping a window from the region of the pattern image containing projected spots, and teaches that the window should encompass spots , indeed "more spots", because the material-coding information resides there. Because each projected spot constitutes a local maximum of the intensity distribution within the pattern image, a window cropped so as to capture one or more spots includes the maximum of the intensity distribution as part of the partial image.
Schick further teaches determining and evaluating individual reflection features within the reflection image, wherein the illumination pattern comprises "at least one point pattern" and the evaluation device analyzes the beam profile of each reflection feature to determine the material property (paras. [0030], [0038]).
It would have been obvious to one of ordinary skill in the art before the effective filing
date of the claimed invention to have determined the maximum of the intensity distribution of the pattern image and cropped the pattern image so as to include that maximum in the partial image. One of ordinary skill in the art would have been motivated to do so because Fang teaches that the material-coding interference information is obtainable "according to the speckle pattern of a spot" and directs that inference proceed "using more spots" (Col. 9). A window positioned to capture the intensity maxima constituted by the projected spots therefore captures the most material-discriminative data available in the image, whereas a window positioned on unilluminated area would capture little or no pattern-derived material information. Positioning the crop on the intensity maxima yields the predictable and desirable result of maximizing the material information content of the partial image while minimizing its size. Determining a local intensity maximum in a digital image is moreover a routine and conventional image processing operation.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
20. Any inquiry concerning this communication or earlier communications from the examiner should be directed to whose telephone number is (571)270-1679. The examiner can normally be reached Monday-Thursday, 6AM - 4PM, PST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Akwasi M Sarpong can be reached at 571.270.3438. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/NEIL R MCLEAN/Primary Examiner, Art Unit 2681
Prosecution Insights