Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
DETAILED ACTION
This action is in response to the applicant's communication filed on 02/20/2026. In virtue of this communication, claims 1-20 are allowed.
Claims 1-2, 8-9, 15-20 have been amended without adding a new subject matter.
Response to Arguments
Applicant's arguments filed on 02/20/2026 with respect to claims 1-20 have been considered.
With regard to prior art rejection, the arguments are not persuasive.
With regard to claim rejection under the 35 USC 101, the rejections have been withdrawn in view of applicant’s argument and amendments.
Applicants Argument : Applicant argued “It can be seen that Zeng discloses the implementation of image feature extraction and object recognition via 24 convolutional layers and 2 fully connected layers in the feature extraction and prediction module, aiming to learn generalized features of the target object for object detection (i.e., detecting whether workers in the construction area wear safety helmets). This is different from the configuration of amended claim 1, which achieves accurate image classification by filtering domain features with low relevance to image classification.”
Zeng fails to disclose or suggest processing the image feature according to the feature filtering layer in the first model to filter the domain feature having a relevance below the first threshold.”
Examiner Answer: Examiner respectfully disagrees, Examiner notes the prior art Zheng ¶[0026] discloses the broadly written claim limitations of image feature including domain features (background) and Class feature (object or target). Further Zhen discloses reducing false detection and filtering out the background and boxes with low scores. examiner notes Zheng ¶[0029] discloses category of target objects in the image can be identified (classified). it uses the non-maximum suppression (NMS) method to filter out the background and boxes with low scores, i.e., boxes with scores below a specified threshold(the background which has score below threshold, low relevancy with regard to foreground). The predicted boxes are output as the detection results. Since the regression and classification of the bboxes are performed directly.
Applicants Argument : Applicant argued “Turning to Gao, Gao only discloses detecting the background of an image and subtracting the background to focus on the foreground. The background detection, background subtraction, and foreground focusing in Gao are all operations at the physical level, rather than the feature subtraction recited in amended claim 1. In addition, Gao does not disclose a dividing feature, nor does it disclose domain features and class features, let alone that the relevance between the domain feature corresponding to the first image and the image classification of the first image is below the first threshold.”
Examiner Answer: Examiner respectfully disagrees, Examiner notes the prior art Gao ¶[0038] discloses subtracting the background which is in this case domain feature from foreground , which is class features for object classification (foreground/object/ class feature.) Examiner notes nowhere in the claim limitations discloses “a dividing feature”. And relevancy threshold comparison limitation Zheng ¶[0029] discloses it uses the non-maximum suppression (NMS) method to filter out the background (domain features)and boxes with low scores, i.e., boxes with scores below a specified threshold(the background which has score below threshold, low relevancy with regard to foreground).
Claim Rejections - 35 USC § 103
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.
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.
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.
Claim(s) 1, 8, 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Zheng et al. (CN113449611A), in view of Gao et al. (US 2019/0114781)
As per claim 1, An image processing method comprising: “obtaining an image feature corresponding to a first image, the image feature comprising a domain feature and a class feature;” (Zheng, ¶[0026] discloses extracting features of input image through feature extraction to obtain feature maps and feature representation of objects (class features) and background (domain features). ¶[0029] discloses inputs the image to be identified into the network model trained in the feature extraction and prediction module. Based on the forward propagation output returned by the feature extraction and prediction module, it obtains the bounding boxes (bboxes) and their categories of the target objects in the image to be identified.)
“processing the image feature according to a feature filtering layer in a first model to obtain a corresponding domain feature of the first image, wherein the feature filtering layer is configured to obtain the domain feature in the image feature;” (Zheng, ¶[0010] discloses the feature extraction network of the network model is the first 20 convolutional layers of the YOLO network. After the feature extraction network, four convolutional layers and two fully connected layers are connected in sequence to form the target recognition branch. ¶[0026] discloses the network structure adopted by the module is based on the GooLeNet model, which includes 24 convolutional layers and 2 fully connected layers. It extracts features through an image pyramid structure and performs feature extraction at different scales through a series of convolutional layers of different scales to obtain more feature maps, learn more generalized feature representations of objects, and is more adaptable to new domains. It has high generalization ability and is highly generalizable. The fully connected layers are used to predict the probability values of image location and category. In order to integrate cross-channel information, a convolutional layer with a 1×1 kernel is used instead of the Inception module of GooLeNet for dimensionality reduction. It can be trained using an entire image to encode the overall category and appearance information of objects, reducing the false detection rate of background and directly optimizing detection performance to improve accuracy.)
“to obtain the class feature of the first image, wherein a relevance between the domain feature and an image classification of the first image is below a first threshold;” (Zheng, ¶[0026] discloses reducing false detection, then ¶[0029] discloses it uses the non-maximum suppression (NMS) method to filter out the background and boxes with low scores, i.e., boxes with scores below a specified threshold, to avoid repeated predictions. The predicted boxes are output as the detection results. Since the regression and classification of the bboxes are performed directly, the running speed is accelerated, and real-time video processing is achieved.)
“and determining an image class of the first image according to the class feature.” (Zheng, ¶[0016] discloses the anchor frame (bbox) is directly regressed and classified, which speeds up the operation and enables real-time video processing. ¶[0029] discloses The predicted boxes are output as the detection results. Since the regression and classification of the bboxes are performed directly, the running speed is accelerated, and real-time video processing is achieved.)
However, Zheng does not explicitly disclose the following which would have been obvious in view of Gao from similar field of endeavor “and subtracting the domain feature from the image feature to obtain the class feature of the first image;”(Gao, ¶[0038] discloses the system 100 can detect and subtract out the background and focus on the foreground (e.g., the defect), which is where the object classification can be performed. The background can be changed repeatedly since the model 104 has been trained to recognize the background, which can be subtracted out of (or removed from) the image by the extraction component 106. Thus, the model 104 can be trained to be robust and not sensitive to background variations.)
Before the effective filing date of the claimed invention it would have been obvious to a person of ordinary skill in the art to combine Gao technique of object classification into Zheng technique to provide the known and expected uses and benefits of Gao technique over target object detection and classification technique of Zheng. The proposed combination would have constituted a mere arrangement of old elements with each performing their known function, the combination yielding no more than one would expect from such an arrangement. Therefore, it would have been obvious to a person of ordinary skill in the art to incorporate Gao to Zheng in order to accurately classifying object with decoupling background from foreground. (Refer to Gao paragraph [0001].)
Claims 8, and 15 have been analyzed and are rejected for the reasons indicated in claim 1 above.
Claim(s) 2-3, 7, 9-10, 14, 16-17, and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Zheng et al. (CN113449611A), in view of Gao et al. (US 2019/0114781), further in view of Jiequan et al. (JP 7376731 B2).
As per claim 2, The method of claim 1, “wherein performing the feature filtering processing on the domain feature in the image feature to obtain the class feature of the first image comprises: “wherein the first model is a model trained according to a plurality of groups of samples” (Gao, ¶[0003] discloses The model can be trained to detect the respective backgrounds with a defined confidence level. The computer executable components can also comprise an extraction component that can employ the model to identify a background of a received image based on the defined confidence level and to decouple a foreground object of the received image based on identification of the background of the received image. ¶[0027] discloses Based on the input data, the training component 102 can train the model 104, and the model 104 can learn to detect the respective backgrounds of the image data 114 with a defined confidence level. The confidence level can be defined based on an acceptable amount of inaccuracy associated with a classification of objects (e.g., foreground objects, defects) in the image data. To train the model 104, the image data 114 can include at least a first image of only a background and at least a second image of an object placed in front of the same background depicted in the first image. For example, the first image can include a background that comprises a lawn and the second image can include a dog standing in the lawn. In another example, the first image can comprise wallpaper as background and the second image can comprise a vase on a table, with the wallpaper as background. The training component 102 can repeatedly train the model 104, such as a classifier (e.g., a predictor found from a classification algorithm), which can become proficient at detecting background since a pure background (e.g., the first image) has been learned by the model 104. Upon or after the model 104 becomes highly trained to detect backgrounds, one or more foreground objects can be placed in the image with the background to facilitate the training. Continuing the lawn and dog example above, a third image can comprise a blanket and a picnic basket on the lawn (e.g., with or without the dog). Further, a fourth image can comprise a vehicle parked on the lawn (e.g., with or without the dog, the blanket, and/or the picnic basket). ¶[0042] discloses to train the model 104 on the image data 114, the training component 102 can perform training on classifiers to detect background images. For example, thousands of images of backgrounds (e.g., image data 114) can be input into the model 104. )
However, Zheng as modified by Gao does not explicitly disclose the following which would have been obvious in view of Jiequan from similar filed of endeavor “where the plurality of groups of samples comprise sample images and sample classes corresponding to the sample images.” (Jiequan ,¶[0017] discloses The sample image set is a data set containing all sample images, and is composed of multiple sample image subsets, each containing sample images of one or more image classes, with the number of image classes contained in each sample image subset being the same, and the total number of images contained in the sample image subsets being different and tending to decrease sequentially. Further ¶[0018] discloses if the sample images include 100 images of image class A, 80 images of image class B, 60 images of image class C, 40 images of image class D, 20 images of image class E, and 10 images of image class F, image classes A and B can form a sample image subset including 180 sample images, image classes C and D can form a sample image subset including 100 sample images, and image classes E and F can form a sample image subset including 30 sample images. ¶[0019] discloses obtaining a large number of sample images from the terminal and classifying the sample images based on the corresponding image types of the sample images. The sample image set may be composed of sample images that conform to a long-tail distribution characteristic (i.e., a large number of images in a small image class and a small number of images in a large image class), or may be composed of sample images that conform to a normal distribution characteristic.)
Before the effective filing date of the claimed invention it would have been obvious to a person of ordinary skill in the art to combine Jiequan technique of image recognition and classification by deep learning into Zheng as modified by Gao technique to provide the known and expected uses and benefits of Jiequan technique over target object detection and classification technique of Zheng as modified by Gao. The proposed combination would have constituted a mere arrangement of old elements with each performing their known function, the combination yielding no more than one would expect from such an arrangement. Therefore, it would have been obvious to a person of ordinary skill in the art to incorporate Jiequan to Zheng as modified by Gao in order to training images with smaller classes and fewer image data for image recognition by neural networks. (Refer to Jiequan paragraph [0003].)
Claims 9 and 16 have been analyzed and are rejected for the reasons indicated in claim 2 above. Additionally, the rationale and motivation to combine the Zheng, Gao and Jiequan references, presented in rejection of claim 2, apply to these claims.
As per claim 3, The method of claim 2, “wherein the first model comprises the feature filtering layer and a first classifier,” (Zheng ¶[0034] discloses The pre-trained classification model consisted of the first 20 convolutional layers, and then an average-pool layer and a fully connected layer were added for pre-training. That is, the fully connected layer was used as the output layer of the pre-trained classification task. )
“wherein the first model is determined by: obtaining a sample image feature and a sample class of a sample image; processing the sample image feature through the feature filtering layer to obtain a sample domain feature; determining a sample class feature according to the sample image feature and the sample domain feature;” (Jiequan, ¶[0016] discloses Step 21: Obtain a sample image set that includes multiple sample image subsets, each containing the same number of image classes. ¶[0017] discloses The sample image set is a data set containing all sample images, and is composed of multiple sample image subsets, each containing sample images of one or more image classes, with the number of image classes contained in each sample image subset being the same, and the total number of images contained in the sample image subsets being different and tending to decrease sequentially. Further see ¶[0018]. ¶[0022] discloses Since multiple branch neural networks are constructed in the image recognition model to be trained, the parameters of the branch neural network can be divided into two parts: shared parameters for extracting common features of the sample images, and individual parameters for extracting sample images of the sample image subset corresponding to the branch neural network based on the shared parameters. ¶[0049] discloses the training module 72 further inputs the sample images into the image recognition model to be trained, such that the base neural network obtains first image features of the sample images, and the branch neural network obtains second image features of the sample images based on the first image features, and determines recognized image classes of the sample images in the set of sample images based on the second image features.) “and updating parameters of the first model according to the sample domain feature, the sample class feature, the sample class, and the first classifier.” (Jiequan, ¶[0004] adjusting model parameters of the image recognition model. ¶[0028-0029] disclose Step 23: Based on the loss value, adjust the model parameters of the image recognition model to be trained until the loss value is lower than a preset threshold, and the image recognition model to be trained is the trained image recognition model.)
Claims 10 and 17 have been analyzed and are rejected for the reasons indicated in claim 3 above.
As per claim 7, The method of claim 3, “wherein the first model further comprises a second classifier, and wherein the method further comprises, after updating the parameters in the feature filtering layer and the first classifier: determining a third loss according to the sample class feature and the sample class; and updating parameters of the second classifier according to the third loss.” (Jiequan, Examiner notes a third loss value interprets as a loss value since the claim depends on claim 3 and there is no preceding loss values. ¶[0023] discloses a second branch neural network corresponding to the second and third sample image subsets of the three sample image subsets. ¶[0025-0026] discloses The loss values of the image recognition model to be trained include a classification loss value and a target classification loss value, where the classification loss value is the loss value for a sample image subset corresponding to the branch neural network of the branch neural network, and the target classification loss value is the loss value for a sample image set of the image recognition model to be trained, and is adjustable. Based on the multiple classification loss values and the target classification loss value, a loss value for training the image recognition model to be trained can be obtained, and the training degree of the entire image recognition model can be determined. ¶[0030] discloses the training module 72 further inputs the sample images into the image recognition model to be trained, such that the base neural network obtains first image features of the sample images, and the branch neural network obtains second image features of the sample images based on the first image features, and determines recognized image classes of the sample images in the set of sample images based on the second image features.)
Claims 14 and 20 have been analyzed and are rejected for the reasons indicated in claim 7 above.
Allowable Subject Matter
Claims 4-6, 11-13, and 18-19 are objected to as being dependent upon a rejected base claim, but would be allowable if i) rewritten in independent form including all of the limitations of the base claim and any intervening claims.
The following is a statement of reasons for the indication of allowable subject matter: the prior art of record, alone or in combination, fails to teach or suggest the limitations set forth by each of claims 4-6, 11-13, and 18-19.
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.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SHAGHAYEGH AZIMA whose telephone number is (571)272-1459. The examiner can normally be reached Monday-Friday, 9:30-6:30.
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, Vincent Rudolph can be reached at (571)272-8243. 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.
/SHAGHAYEGH AZIMA/Examiner, Art Unit 2671
Prosecution Insights