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 . It is responsive to the submission dated 09/10/2024. Claims 1-20 are presented for examination. Claims 1, 7 and 14 are independent claims.
Information Disclosure Statement
2. The information disclosure statements (IDSs) submitted on 09/10/2024 are in compliance with the provisions of 37 CFR 1.97 and are being considered by the Examiner.
Claim Objections
3. Claims 1, 7 and 14 are objected to because of the following informalities: in line 9 of said claims, replace “vector image generator model based” with - vector image generator model is based-. Appropriate correction is required.
Duplicate Claims Warning
4. Applicant is advised that should claim 1 be found allowable, claim 7 will be objected to under 37 CFR 1.75 as being a substantial duplicate thereof. When two claims in an application are duplicates or else are so close in content that they both cover the same thing, despite a slight difference in wording (e.g., translate received image to raw image to generate an augmented image), it is proper after allowing one claim to object to the other as being a substantial duplicate of the allowed claim. See MPEP § 608.01(m).
Claim Rejections - 35 USC § 101
5. 35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
6. Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more.
Independent claim 1 recites the followings:
1. “A training method for a vector image generator model, the method comprising:
receiving an image; generating an augmented image by transforming a raw image included in the received image; inputting at least one of the raw image or the augmented image into the vector image generator model; and outputting a vector image using the vector image generator model, the vector image output by the vector image generator model based on the raw image and the augmented image, wherein the vector image generator model comprises a neural network model, and wherein the vector image generator model is configured to output the raw image as the vector image output from the vector image generator model when the raw image and the augmented image are input to the vector image generator model.”
Under step 1, it is determined whether the claims are directed to a statutory category of invention (see MPEP 2106.03(II)). In the instant case, while the claim fall within statutory categories, under revised Step 2A, Prong 1 of the eligibility analysis (MPEP 2106.04), claim 1 recites an abstract idea of “Mental Processes” of gathering information in the mind and output it on paper.
For instances, the combination of steps, as drafted in claim 1, are processes that, under their broadest reasonable interpretations, cover performances of mental processes, such as performing concepts in the mind or using a pen and paper, but for the recitation of training a vector image generator model including a neural network model.
For examples, the image receiving step and the augmented image generating step are broadly construed as a person/user, upon accessing a physical image drawing book, mentally or visually analyzed a specific annotation of an object drawn within an image in the image drawing book for selection. The image inputting step is broadly construed as the user, after performing the mental evaluation and selection, uses a pen and paper to manually sketch a replica of the annotated object image by connecting dots together to form an illustration (e.g., transforming a raw image included in the received image). The outputting of the vector image step is the result of the sketch image drawn on the paper, which by choice, can be printed, scanned or photographed by the user for further evaluation.
Thus, claim 1 recites the abstract idea of gathering and grouping “mental processes”.
Under revised Step 2A, Prong 2 of the eligibility analysis, if it is determined that the claims recite a judicial exception, it is then necessary to evaluate whether the claims recite additional elements that integrate the judicial exception into a practical application of that exception. In this case, representative claim 1 is not integrated into a practical application, as it only recites the additional elements of training a vector image generator model and a neural network model.
These additional elements individually and in combination are recited at a high level of generality such that they amount to no more than mere instruction to apply the exception using generic computer components. See: MPEP 2106.05(f). These elements appear to be mere tangential additions to the abstract idea(s) and amount to extra-solution activity concerning mere data gathering. The addition of an insignificant extra-solution activity limitation does not impose meaningful limits on the claim such that is it not nominally or tangentially related to the invention.
In the claimed context, the use of training a vector image generator model and a neural network model for outputting the image only presents the idea of a solution, while failing to describe how the image generator model and neural network model are used to achieve the solution of outputting the raw image. The added elements appear to be mere tangential addition to the abstract idea(s) and amount to extra-solution activity concerning mere data gathering, evaluation and outputting. They do not impose meaningful limits on the claim such that they are not nominally or tangentially related to the invention. Accordingly, the additional element of training a vector image generator model including a neural network model does not integrate the abstract idea into a practical application of the invention.
Under Step 2B of the eligibility analysis, if it is determined that the claims recite a judicial exception that is not integrated into a practical application of that exception, it is then necessary to evaluate the additional elements individually and in combination to determine whether they provide an invention concept (i.e., whether the additional elements amount to significantly more than the exception itself), as discussed in MPEP 2106.05.
The judicial exception is not integrated into a practical application, because neural network or image generator model (as the additional elements) are merely incidental or token additions to the claim that do not alter or affect how the process steps or functions in the abstract idea(s) are performed, by merely providing an output of the recited mental process. Therefore, the training a vector image generator model comprising a neural network model is an insignificant extra-solution activity limitation that merely provides an output to the recited mental processes. They do not impose any meaningful limits on practicing the mental processes beyond a general linking to a technological environment. See: MPEP 2106.05(h). Generator and/or neural network models, in the field artificial intelligence or computer visions, are well-known, routine, and conventional such as to not qualify as inventive concepts.
As such, claim 1 is not patent eligible under 35 USC 101.
Both of independent claims 7 and 14 recite features that are substantially the same in context and scope to those of claim 1. Thus, claims 7 and 14 are not patent eligible under 35 USC 101 for the same reasons stated above with respect to claim 1.
Dependent claims 2 and 3 narrow the claimed abstract ideas by defining processing procedures that a user with a pen and paper can use to apply certain modifications on the drawn/sketched image using.
Dependent claims 4-6 further narrow the claimed abstract ideas by reciting steps further define how the user can output the image (e.g., by training the vector image generator model via a self-supervised learning method, by using a plurality of convolution layers and/or by means of a variational autoencoder). The additional elements of training the vector image generator model via a self-supervised learning method, by using a plurality of convolution layers and/or by means of a variational autoencoder are merely incidental or token additions to the claims that do not alter or affect how the process steps or functions in the abstract idea(s) are performed, by merely providing an output of the recited mental process. Therefore, the use of a self-supervised learning method, including using a plurality of convolution layers and/or a variational autoencoder are insignificant extra-solution activities that merely provide an output to the recited mental processes. They do not impose any meaningful limits on practicing the mental processes beyond a general linking to a technological environment. See: MPEP 2106.05(h). Training a generator model as a self-supervised learning method, or by using a plurality of convolution layers and/or by means of a variational autoencoder, in the field artificial intelligence or computer visions, are well-known, routine, and conventional such as to not qualify as inventive concepts. These steps add no meanings to the abstract ideas.
Claims 8-10 and 15-16, like claims 2-3, fail to remedy the deficiencies of parent claims 7 and 14 above. As claimed, they merely provide further details the claimed elements shown above to be abstract without significantly more. These claims are therefore rejected under 35 USC 101 for at least the same rationale as applied to claims 2-3 and their parent claims above and incorporated herein.
Claims 11-13 and 17-20, like claims 4-6, fail to remedy the deficiencies of parent claims 7 and 14 above. As claimed, they merely provide further details the claimed elements shown above to be abstract without significantly more. These claims are therefore rejected under 35 USC 101 for at least the same rationale as applied to claims 4-6 and their parent claims above and incorporated herein.
As such, it is submitted that neither of claims 1-20 or the combination thereof are patent eligible under 35 USC 101.
Claim Rejections - 35 USC § 112
7. The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
8. Claims 1-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 1, for example, recites:
“A training method for a vector image generator model, the method comprising:
receiving an image; generating an augmented image by transforming a raw image included in the received image; inputting at least one of the raw image or the augmented image into the vector image generator model; and outputting a vector image using the vector image generator model, the vector image output by the vector image generator model based on the raw image and the augmented image, wherein the vector image generator model comprises a neural network model, and wherein the vector image generator model is configured to output the raw image as the vector image output from the vector image generator model when the raw image and the augmented image are input to the vector image generator model.”
The cited limitations as drafted render the claim indefinite because they provide no concrete functional or structural features explaining how the processes of said method are implemented. These steps, as claimed, appear to be a concatenation of a block box experiment, of which only inputs and outputs are specified.
Particularly, it is unclear as to what is/are doing the receiving, the generating, the transforming, the inputting and the outputting steps. Are these steps automated or they involve the gathering of information in the mind of a person?
Further, while the claim appears to cite using “a vector image generator model comprises a neural network model” to train the “raw image or the augmented image”, the limitation(s) that follows the citing fails to provide how the image generator model and/or the neural network model is trained to achieve the stated solution. Typically, using a generator model and/or a neural network requires some specialized hardware for configuring the model datasets during training to determine how it will interact with the input and output of data. In the present claim, no such detail is provided explaining how the model(s) is being trained. Merely stating that the neural network of the generator model is a processor of an electronic device, as cited in the original disclosure at paras. 75-78, is not detailed enough to make use of the invention. As such, the limitations fail to limit the claim.
Claim 3 lacks sufficient antecedent basis for the “plurality of the augmented images”.
Independent claims 7 and 14 recite limitations that are similar in scope to those of claim 1. Claims 7 and 14 are indefinite for the same reasons as claim 1.
Claims 9 and 16 are indefinite for the same reasons as claim 3.
The claims not specifically cited in this rejection are rejected as being dependent upon their rejected base claims.
Claim Rejections - 35 USC § 103
9. 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.
10. Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over the article to Cao et al. entitled “SVGformer: Representation Learning for continuous vector graphics using transformers”.
Considering, Cao disclose training method for a vector image generator model (e.g., using a transformer-based model to learn vector graphics; see abstract), the method comprising:
generating an augmented image by transforming a raw image included in a received image (for example, Cao discloses transforming an image structured in bitmap format to vector graphics format and yield the representation of vector objects; see the “Introduction” section of Cao, wherein the vector object representation corresponds to the generated “augmented image”. See also fig. 1 of Cao for further describing the input image to encompass a raw image);
inputting at least one of the raw image or the augmented image into the vector image generator model (for example, Cao discloses our proposed model is the first to explicitly consider vector geometric information as well as directly deal with the raw input of SVG format in an end-to-end encoder-decoder fashion…….. SVGformer captures both geometric information and curve semantic information with the geometric self-attention module, which synergizes the strengths of MAT and the transformer-based neural network to handle long-term sequence relationships in SVG. See top-left column at page 10094 of section 1 of Cao. Section 4.2.1 of Cao further describes the SVGformer model to reconstruct more details of the raw image of the vector graphics to be used as input to the model to obtain the latent representation of a given SVG image (e.g. augmented image)); and
outputting a vector image using the vector image generator model, the vector image output by the vector image generator model is based on the raw image and the augmented image, wherein the vector image generator model comprises a neural network model (e.g., Im2Vec [25] proposes a new neural network that can generate complex vector graphics with varying topologies and only requires indirect supervision from readily-available raster training images (i.e., with no vector counterparts). Instead of directly using bitmap format, DeepVecFont [33] introduces a new generative paradigm to handle unstructured data (e.g., vector glyphs) by randomly sampling plausible synthesis results and a dual-modality learning strategy that utilizes both image-aspect and sequence-aspect features of fonts to synthesize vector glyphs. See page 2nd para., right-column of page 10094).
Cao does not explicitly describe the vector image generator model is configured to output the raw image as the vector image output from the vector image generator model when the raw image and the augmented image are input to the vector image generator model.
Cao, however, discloses: We propose a transformer-based deep neural network, SVGformer, as a general solution for multiple vector graphics downstream tasks. The overall architecture of SVG former is illustrated in Figure 2. The rawt-th SVG’s arguments contain the type and corresponding coordinates of its commands, where the coordinates with continuous values are first constructed into a tensor format… then the input tensor is fed into the embedding layer to get the input vector for the encoder-decoder model. The learnable embedding layer contains three components to specifically abstract the information from the original input:1) the continuous token component uses a 1D convolution to map Xt into an embedding vector Ut; 2) the position component captures the positional information Pt from Xt; 3) the geometric component treats the outline labels li (refer to Figure 1) as a discrete token to map extra contextual geometric information St to the embedding vector. In addition, we build an adjacency matrix E based on the labels and the neighboring commands in order to further utilize the structural information in the training process. Next, we combine the embedding vectors to feed the encoder module, where we introduce the geometric self-attention to replace the canonical self-attention in order to generate a compact representation with a higher receptive field efficiently. See section 3.1 at page 10095 of Cao. In addition, section 3.3 of Cao also teaches that the representation learning requires the model to learn the inductive biases from the data and training task which can facilitate the downstream task. Thus, any inductive bias from raw input is able to coach the training process. In this work, we have identified two inductive biases present in raw input: continuous values and geometric information. While previous related works have typically relied on manually discretizing continuous values, our model is able to support continuous-valued inputs.
Therefore, considering that in Cao the transformer-based deep neural network, SVGformer model is able to generate vector image objects by detecting raw image from bitmap image and input the raw image with biases that supports continuous values and geometric information to the SVG model for training multiple vector graphics downstream tasks; inputting the detected raw image (that represents the vector objects as augmented image) from the bitmap image to the SVG model to output the raw image as the vector image when the raw image and the augmented image are input to the vector image generator model, would be an obvious variant of the continuous vector graphics downstream task result of Cao.
Thus, the artisan skill in the art, given the teachings of Cao, before the effective filing date of the claimed invention, would have found it obvious to further modify the transformer-based SVGformer model of Cao to cause said model during training to output the raw image as the vector image when the raw image detected in the original bitmap image and the transformed vector objects (as augmented image) from the converted bitmap image are input to the transformer-based deep neural network, SVGformer model for training. The advantage to modify the Cao teachings as such would have been to allow efficient computation of the overall architecture of the SVGformer model; in order to generate a compact representation with a higher receptive field efficiently. See section 3.1 at page 10095 of Cao.
As per claim 2, Cao discloses transforming the raw image comprises at least one of a rotation transformation, a symmetry transformation, a size transformation, or a translation (e.g., mapping extra contextual geometric information to the embedding vector model for training by utilizing positional information to transform geometric information from raw input image data, or labelling geometric segment information or transforming input tensor datasets and/or by performing continuous feature embedding. See fig. 2 and sections 3.1 to 3.3 of Cao. See also table 2 and section 4.2.1).
As per claim 3, Cao discloses a plurality of the augmented images are input to the vector image generator model (for examples, Cao teaches an original image in bitmap format is transformed to scalable vector graphics format for vector images represented as vector objects attributes (e.g., augmented images) to be trained by the transformer-based SVGformer model, by considering raw data detected in the bitmap image as input to the model. See “introduction” section and sections 1 and 3.1 of Cao).
As per claim 4, Cao discloses the vector image generator model comprises a plurality of convolution layers (e.g., the learnable embedding layer contains three components (e.g., convolution 1D embedding layer, Position embedding layer and geometric embedding layer) to specifically abstract the information from the original input image. See fig. 2 and sections 3.1 to 3.3 of Cao). Section 3.4, page 10097 of Cao further teaches Our model can produce command-level output Ageo ∈ RN× d with L multiple layers GCN where l th layer output Hl can be used to represent the latent vectors to get the final output).
As per claim 5, Cao discloses the vector image generator model comprises a variational autoencoder (e.g., the transformer-based SVGformer model uses a deep learning neural network to train vector graphics by combining embedding vectors to feed an encoder module for encoding geometric segment information of the vector graphics images, such that embedding layer can fully utilize the translation equivariance of convolutional operation for SVG input and can have the ability of rotation invariance and size invariance with data augmentation. See fig. 2 and section 3.3 of Cao). See also section 2 (e.g., Representation learning for SVG) of Cao that teaches a generative model with an autoencoder.
As per claim 6, Cao discloses the vector image generator model is learned in a self-supervised learning method (e.g., the encoder-decoder architecture of the transformer model utilizes a self-attention mechanism. See fig. 2 and section 2 (e.g., Representation learning for SVG) of Cao and also section 3.4).
The invention of claim 7 contains features that correspond in scope with the limitations recited claim 1. As the limitations of claim 1 were found obvious over the teachings of Cao, it is readily apparent that the applied prior art performs the underlying elements. As such, the limitations of claim 7 are, therefore, subject to rejections under the same rationale as claim 1. In addition, Cao discloses transforming a format of the received image and generating a raw image that is a vector image to generate an augmented image (for examples, Cao teaches an original image in bitmap format is transformed to scalable vector graphics format for vector images represented as vector objects attributes (e.g., augmented images) to be trained by the transformer-based SVGformer model, by considering raw data detected in the bitmap image as input to the model. See “introduction” section and sections 1 and 3.1 of Cao. Section 4.2.1 of Cao further describes the SVGformer model to reconstruct more details of the raw image of the vector graphics to be used as input to the model to obtain the latent representation of a given SVG image (e.g. augmented image)).
As per claim 8, Cao discloses the transforming the raw image comprises at least one of 90° rotation transformation, 180° rotation transformation, 270° rotation transformation, X-axis symmetry transformation, Y-axis symmetry transformation, a combination of X-axis symmetry transformation and 90° rotation transformation, a combination of X-axis symmetry transformation and 180° rotation transformation, a combination of X-axis symmetry transformation and 270° rotation transformation, a combination of Y-axis symmetry transformation and 90° rotation transformation, a combination of Y-axis symmetry transformation and 180° rotation transformation, and a combination of Y-axis symmetry transformation and 270° rotation transformation (for example, Cao teaches performing position embedding layer can fully utilize the translation equivariance of convolutional operation for SVG input and can have the ability of rotation invariance and size invariance with data augmentation. Section 3.3 page 10096, where the translation by rotation invariance obviously encompasses at least one of the 180 or 270° rotation transformation, and/or the X-axis or Y-axis symmetry transformations, as claimed).
Claim 9 is rejected under the same rationale as claim 2.
As per claims 10-11, Cao discloses the received image comprises at least one of a vector image format and a raster graphics format, wherein the vector image has a scalable vector graphics (SVG) format. See “introduction” section and sections 1 and 3.1 of Cao.
As per claim 12, Cao discloses the vector image generator model comprises a encoder and a decoder (see fig. 2 and sections 3.4 and 3.5 of Cao) and variational autoencoder (See section 2 (e.g., Representation learning for SVG) of Cao that teaches a generative model with an autoencoder).
As per claim 13, Cao discloses at least one of the encoder and the decoder comprises at least one of a recurrent neural network (RNN), a long short-term memory (LSTM) network, or a transformer network (e.g., Cao discloses SVGformer-no-geo aborts the geometric information inside the self-attention mechanism, which indicates the importance of structural inductive bias for encoder-decoder-based transformer models (e.g., transformer network). See section 4.3 at page 10099 of Cao. Cao further teaches proposing a transformer-based deep neural network, SVGformer model, as a general solution for multiple vector graphics downstream tasks. See section 3.1, as the basis for training the encoder-to-decoder model using a recurrent neural network and/or long short-term memory neural network).
The invention of claim 14 contains features that correspond in scope with the limitations recited claim 1. As the limitations of claim 1 were found obvious over the teachings of Cao, it is readily apparent that the applied prior art performs the underlying elements. As such, the limitations of claim 14 are, therefore, subject to rejections under the same rationale as claim 1. In addition, Cao discloses inputting labeled input data to the vector image generator model (for examples, Cao teaches an original image in bitmap format is transformed to scalable vector graphics format for vector images represented as vector objects attributes (e.g., augmented images) to be trained by the transformer-based SVGformer model, by considering raw data detected in the bitmap image as input to the model. See “introduction” section and sections 1 and 3.1 of Cao. Section 3.2 of Cao discloses the SVG former model performing geometric information from raw input by obtaining semantic labeling of segments with MAT. This tool leverages the medial axis and outlines information to find new segments labeled based on different geometric relationships).
As per claim 15, Cao discloses transforming the received image into the vector image (for examples, Cao teaches an original image in bitmap format is transformed to scalable vector graphics format for vector images represented as vector objects attributes (e.g., augmented images) to be trained by the transformer-based SVGformer model, by considering raw data detected in the bitmap image as input to the model. See “introduction” section and sections 1 and 3.1 of Cao. Section 4.2.1 of Cao further describes the SVGformer model to reconstruct more details of the raw image of the vector graphics to be used as input to the model to obtain the latent representation of a given SVG image (e.g. augmented image)).
As per claim 16, Cao discloses label data of the input data indicates whether the input data is the raw image, and includes a transformation method for the raw image when the input data is the augmented image (for examples, Cao teaches an original image in bitmap format is transformed to scalable vector graphics format for vector images represented as vector objects attributes (e.g., augmented images) to be trained by the transformer-based SVGformer model, by considering raw data detected in the bitmap image as input to the model. See “introduction” section and sections 1 and 3.1 of Cao. At Section 3.2, Cao further discloses the SVG former model performing geometric information from raw input by obtaining semantic labeling of segments with MAT. This tool leverages the medial axis and outlines information to find new segments labeled based on different geometric relationships. Additionally, Section 4.2.1 of Cao describes the SVGformer model to reconstruct more details of the raw image of the vector graphics to be used as input to the model to obtain the latent representation of a given SVG image (e.g. augmented image)). As such, Cao teaches the features of claim 16.
As per claims 17-18, Cao discloses the vector image generator model comprises an encoder and a decoder, and label data is input to each of the encoder and the decoder (see fig. 2), wherein label data input to the encoder is same as label data input to the decoder (for examples, Cao teaches the end-to-end encoder-decoder performing Geometric Dependency in Self-Attention by determining Segments with the same label to share a strong relationship even if they are at a long distance in the input sequence, which cannot be reflected in the original self-attention module which only considers the semantic characteristics. In addition, commands can be represented as the nodes lying in a non-Euclidean space and linked by edges with the geometric information. such as the adjacency and the same segment la bel. In this work, we propose to apply the graph convolution network (GCN) [15] to extract the structural relationship between different commands. Specifically, we first use the geometric label from Section 3.2 to build the weight matrix E where the commands adjacent to each other or sharing the same label have an edge in the weight matrix E, otherwise there is no edge in the weight matrix E. see section 3.4, para. 3 page 10097 and the end of section 4.3, page 100100 of Cao).
As per claim 19, Cao discloses label data input to the encoder is different from label data input to the decoder (e.g., Cao discloses in the SVG former model the decoder part is modified from the decoder structure in [41], which can reconstruct the long sequences of SVG at one forward operation rather than a step-by-step way with the input. See section 3.5. and, for the loss function, we need to reconstruct the input commands as well as predict the type of the commands. Thus, we choose the reconstruction loss as the mean squared error (MSE) loss function between the reconstructed commands and the ground-truth command and the classification loss as the cross entropy (CE) with logits between the predicted command type and the type label. See section. 3.6). As such, it is submitted that the Cao reference obviously encompasses that label data input to the encoder is different from label data input to the decoder. See also “Classification Task”, section 4.2.1 section of Cao.
As per claim 20, Cao discloses the vector image generator model is learned in a self-supervised learning method comprising label data (e.g., the encoder-decoder architecture of the transformer model utilizes a self-attention mechanism and use the geometric label from Section 3.2 to build the weight matrix E where the commands adjacent to each other or sharing the same label have an edge in the weight matrix E See fig. 2 and section 3.4).
Conclusion
11. The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Caron et al. (US 20210250565) discloses a method and apparatus may be used for performing a scene-based automatic white balance correction. The method may include obtaining an input image. The method may include obtaining a raw image thumbnail. The method may include obtaining an augmented image thumbnail. The method may include computing a histogram from an image thumbnail. The method may include determining a scene classification. The method may include learning a filter. The filter may be learned from one or several different instances of the raw image thumbnail, the augmented image thumbnail, the scene classification, or any combination thereof. The method may include applying the filter to the histogram to determine white balance correction coefficients and obtain a processed image.
12. Any inquiry concerning this communication or earlier communications from the examiner should be directed to WESNER SAJOUS whose telephone number is (571)272-7791. The examiner can normally be reached on M-F 9:30 TO 6:30.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Broome Said can be reached on 571-272-2931. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/WESNER SAJOUS/Primary Examiner, Art Unit 2612
WS
03/03/2026