Detailed Action
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 .
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1-2, 4, 7, 9, 11-14, 16, 18, and 20 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Esmaeilzehi et al. (“MuRNet: A deep recursive network for super resolution of bicubically interpolated images”, Signal Processing: Image Communication, 2021), (hereinafter Esmaeilzehi).
Regarding claim 1, Esmaeilzehi teaches an image processing system including:
a processing subsystem that includes one or more processors and one or more memories coupled with the one or more processors (Esmaeilzehi, “The Keras [38] library and TensorFlow package [39] are used for implementing MuRNet. The training of MuRNet is carried out on the machine with Intel Core i7 CPU @4.2 GHz, 16-GB RAM and GPU Nvidia Titan X (Pascal).”, pg. 5, 1st column, 1st full paragraph), the processing subsystem configured to cause the system to:
receive a lower-resolution image; generate a first higher-resolution image by applying spatial interpolation to the lower-resolution image (Esmaeilzehi, “In this section, the idea behind the proposed scheme for single image super resolution is presented and its architecture is developed. The training details of the network and hyperparameters used by it are also explained.”, pg. 3, 1st column, 1st paragraph, “The overall architecture of the proposed network for image super resolution is shown Fig. 1. The bicubic interpolated low resolution image, whose spatial resolution is the same as that of the ground truth, is fed to the proposed network.”, pg. 3, 1st column, 2nd paragraph, lines 1-4, see Fig. 1, Bicubic LR Image, MuRNet is proposed to perform single image super resolution for bicubic interpolated images. The training includes generating low resolution images by applying bicubic down sampling to training images. This low-resolution image is then upscaled using bicubic interpolation to create an input image for the network at the target resolution.);
generate a refinement layer by applying a neural network to the lower-resolution image, the neural network trained to predict a residue in the first higher-resolution image (Esmaeilzehi, “The features of the bicubic interpolated image are extracted using a convolutional layer, which employs 64 convolutional filters each of size 7 × 7 × 1. These feature maps are then processed by a recursive block to be developed and explained in the next subsection. Use of a recursive network would increase the nonlinear mapping capability of the image super resolution scheme, and therefore, would result in a better estimation of the high resolution image, if a sufficiently large number of recursions is used, while keeping the number of parameters of the network unchanged. However, since the effective depth of the network increases as the number of recursions is increased, the gradient vanishing problem of the network would appear thus hindering its learning process. To address this problem, a global residual skip connection is used in the proposed scheme, through which the residue between the ground truth and bicubic interpolated image is learnt by the deep network.”, pg. 3, 1st column, 2nd paragraph, lines 4-18, see Figs. 1 and 2, The network is then trained on these inputs to predict a residual image, which represents the high-frequency details lost during interpolation. This training generates the network which serves as a “refinement layer”.); and
generate a second higher-resolution image by refining the first higher-resolution image using the refinement layer (Esmaeilzehi, “The output feature maps of the recursive block after completing all the recursions are fed to a convolutional layer, that employs a single convolutional filter of size 7 × 7 × 64, to obtain the residual image. Finally, the residual image is added to the bicubic interpolated image and the estimated high resolution image is yielded.”, pg. 3, 1st column, 2nd paragraph, lines 19-23, see Fig. 1, HR image, The predicted residual image can then be combined with the bicubic LR image to produce a final high-resolution output.).
Regarding claim 2, Esmaeilzehi teaches the image processing system of claim 1, wherein the spatial interpolation includes at least one of a bicubic interpolation, a linear interpolation, or a Lanczos interpolation (Esmaeilzehi, “The overall architecture of the proposed network for image super
resolution is shown Fig. 1. The bicubic interpolated low resolution image, whose spatial resolution is the same as that of the ground truth, is fed to the proposed network.”, pg. 3, 1st column, 2nd full paragraph, lines 1-4).
Regarding claim 4, Esmaeilzehi teaches the image processing system of claim 1, wherein the neural network includes a convolutional neural network (Esmaeilzehi, “The features of the bicubic
interpolated image are extracted using a convolutional layer, which employs 64 convolutional filters each of size 7 × 7 × 1. These feature maps are then processed by a recursive block to be developed and explained in the next subsection.”, pg. 3, 1st column, 2nd full paragraph, lines 4-8).
Claim 7 corresponds to claim 1, reciting a computer-implemented method to perform the steps according to claim 1. Esmaeilzehi teaches a computer-implemented method to perform the steps according to claim 1 (Esmaeilzehi, “The Keras [38] library and TensorFlow package [39] are used for implementing MuRNet. The training of MuRNet is carried out on the machine with Intel Core i7 CPU @4.2 GHz, 16-GB RAM and GPU Nvidia Titan X (Pascal).”, pg. 5, 1st column, 1st full paragraph, see Figs. 1 and 2). As indicated in the analysis of claim 1, Esmaeilzehi teaches all of the limitations according to claim 1. Therefore, claim 7 is rejected for the same reason as claim 1.
Regarding claim 9, Esmaeilzehi teaches the computer-implemented method of claim 7, wherein said refining includes summing the first higher-resolution image and the residue refinement layer (Esmaeilzehi, “Finally, the residual image is added to the bicubic interpolated image and the estimated high resolution image is yielded.”, pg. 3, 1st column, 2nd full paragraph, lines 22-23, see Fig. 1).
Regarding claim 11, Esmaeilzehi teaches the computer-implemented method of claim 7, further comprising training the neural network (Esmaeilzehi, “In this subsection, the training procedure of MuRNet is described. All of the convolutional layers in the recursive block are followed by ReLU [34] activations.”, pg. 4, 1st column, 3rd full paragraph).
Regarding claim 12, Esmaeilzehi teaches the computer-implemented method of claim 11, wherein said training includes downscaling a high-resolution image (Esmaeilzehi, “All the images are divided into 30 363 sub-images each of size 48 × 48 to produce the training samples. In order to generate the degraded low resolution images, the bicubic downsampling operation is applied to the original high resolution images.”, pg. 4, 1st column, 5th full paragraph, lines 5-8, The training includes generating low resolution images by applying bicubic down sampling to high-resolution training images).
Regarding claim 13, Esmaeilzehi teaches the computer-implemented method of claim 12, wherein said training includes apply spatial interpolation to upscale the downscaled image. (Esmaeilzehi, “In this section, the idea behind the proposed scheme for single image super resolution is presented and its architecture is developed. The training details of the network and hyperparameters used by it are also explained.”, pg. 3, 1st column, 1st paragraph, “The overall architecture of the proposed network for image super resolution is shown Fig. 1. The bicubic interpolated low resolution image, whose spatial resolution is the same as that of the ground truth, is fed to the proposed network.”, pg. 3, 1st column, 2nd paragraph, lines 1-4, see Fig. 1, Bicubic LR Image, The low-resolution image is then upscaled using bicubic interpolation to create an input image for the network at the target resolution.)
Regarding claim 14, Esmaeilzehi teaches the computer-implemented method of claim 7, wherein the neural network includes a convolutional neural network (Esmaeilzehi, “The features of the bicubic interpolated image are extracted using a convolutional layer, which employs 64 convolutional filters each of size 7 × 7 × 1. These feature maps are then processed by a recursive block to be developed and explained in the next subsection.”, pg. 3, 1st column, 2nd full paragraph, lines 4-8).
Claim 16 corresponds to claim 1, with the addition of a non-transitory computer-readable medium or media having stored thereon machine interpretable instructions which, when executed by a processing system, cause the processing system to perform the functions according to claim 1. Esmaeilzehi teaches the addition of a non-transitory computer-readable medium or media having stored thereon machine interpretable instructions (Esmaeilzehi, “The Keras [38] library and TensorFlow package [39] are used for implementing MuRNet. The training of MuRNet is carried out on the machine with Intel Core i7 CPU @4.2 GHz, 16-GB RAM and GPU Nvidia Titan X (Pascal).”, pg. 5, 1st column, 1st full paragraph). As indicated in the analysis of claim 1, Esmaeilzehi teaches all the limitation according to claim 1. Therefore, claim 16 is rejected for the same reason as claim 1.
Regarding claim 18, Esmaeilzehi teaches the non-transitory computer-readable medium or media of claim 16, wherein said refining includes summing the first higher-resolution image and the residue refinement layer (Esmaeilzehi, “Finally, the residual image is added to the bicubic interpolated image and the estimated high resolution image is yielded.”, pg. 3, 1st column, 2nd full paragraph, lines 22-23, see Fig. 1).
Regarding claim 20, Esmaeilzehi teaches the non-transitory computer-readable medium or media of claim 16, wherein the instructions further cause the processing system to train the neural network (Esmaeilzehi, “In this subsection, the training procedure of MuRNet is described. All of the convolutional layers in the recursive block are followed by ReLU [34] activations.”, pg. 4, 1st column, 3rd full paragraph).
Claim Rejections - 35 USC § 103
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.
Claims 3, 6, 8, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Esmaeilzehi et al. (“MuRNet: A deep recursive network for super resolution of bicubically interpolated images”, Signal Processing: Image Communication, 2021) in view of Xiang et al. (“Zooming Slow-Mo: Fast and Accurate One-Stage Space-Time Video Super-Resolution”, Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020.), (hereinafter Xiang).
Regarding claim 3, Esmaeilzehi teaches the image processing system of claim 1. Esmaeilzehi does not teach wherein the spatial interpolation includes spatiotemporal interpolation.
However, Xiang teaches wherein the spatial interpolation includes spatiotemporal interpolation (Xiang, “In this paper, we explore the space-time video super-resolution task, which aims to generate a high-resolution (HR) slow-motion video from a low frame rate (LFR), low-resolution (LR) video.”, see abstract, lines 1-4, “we propose to learn a feature temporal interpolation function f(·) to directly synthesize the intermediate feature map
F
2
L
(see Fig. 3). A general form of the interpolation function can be formulated as: (see eq. (1))”, pg. 3373, 1st column, 1st full paragraph, lines 7-10, see Figs. 2 and 3).
Esmaeilzehi teaches applying bicubic interpolation to low-resolution images to train a network to predict residual images for image super-resolution (Esmaeilzehi, pg. 3, 1st column, 2nd paragraph, see Figs. 1 and 2). Esmaeilzehi does not teach wherein the interpolation is spatiotemporal. Xiang teaches performing super resolution by implementing video frame interpolation which considers spatial and temporal features (see above). Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to have modified the interpolation of Esmaeilzehi to include spatiotemporal interpolation applied for video frames as taught by Xiang (Xiang, pg. 3373, 1st column, 1st full paragraph, lines 7-10, see Figs. 2 and 3). The motivation for doing so would have been extend Esmaeilzehi’s image super-resolution to video processing, thereby generating visually appealing high-resolution videos by incorporating spatiotemporal interpolation to consider the additional temporal features of video frames (as suggested by Xiang, “Space-Time Video Super-Resolution (STVSR) [30] aims to automatically generate a photo-realistic video sequence with a high space-time resolution from a low-resolution and low frame rate input video. Since HR slow-motion videos are more visually appealing containing fine image details and clear motion dynamics, they are desired in rich applications, such as film making and high-definition television.”, pg. 3370, 2nd column, 1st full paragraph). Further, one skilled in the art could have combined the elements as described above by known methods with no change in their respective functions, and the combination would have yielded nothing more than predictable results. Therefore, it would have been obvious to combine the teachings of Esmaeilzehi with Xiang to obtain the invention according to claim 3.
Regarding claim 6, Esmaeilzehi teaches the image processing system of claim 1. Esmaeilzehi does not teach wherein the image is a frame of video.
However, Xiang teaches wherein the image is a frame of video (Xiang, “Given an LR, LFR video sequence… our goal is to generate the corresponding high-resolution slow-motion video sequence…To fast and accurately increase resolution in both space and time domains, we propose a one-stage space-time super-resolution framework: Zooming Slow-Mo as illustrated in Figure 2. The framework mainly consists of four parts: feature extractor, frame feature temporal interpolation module, deformable ConvLSTM, and HR frame reconstructor.”, pg. 3372, 2nd column, 1st full paragraph).
Esmaeilzehi teaches applying bicubic interpolation to low-resolution images to train a network to predict residual images for image super-resolution (Esmaeilzehi, pg. 3, 1st column, 2nd paragraph, see Figs. 1 and 2). Esmaeilzehi does not teach wherein the images are frames of video. Xiang teaches performing super resolution on sequences of video frames to generate high-resolution videos (see above). Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to have modified the network of Esmaeilzehi to be applied for video frames as taught by Xiang (Xiang, pg. 3372, 2nd column, 1st full paragraph, see Figs. 2 and 3). The motivation for doing so would have been extend Esmaeilzehi’s image super-resolution to video processing, thereby generating visually appealing high-resolution videos (as suggested by Xiang, “Space-Time Video Super-Resolution (STVSR) [30] aims to automatically generate a photo-realistic video sequence with a high space-time resolution from a low-resolution and low frame rate input video. Since HR slow-motion videos are more visually appealing containing fine image details and clear motion dynamics, they are desired in rich applications, such as film making and high-definition television.”, pg. 3370, 2nd column, 1st full paragraph). Further, one skilled in the art could have combined the elements as described above by known methods with no change in their respective functions, and the combination would have yielded nothing more than predictable results. Therefore, it would have been obvious to combine the teachings of Esmaeilzehi with Xiang to obtain the invention according to claim 6.
Regarding claim 8, Esmaeilzehi teaches the computer-implemented method of claim 7. Esmaeilzehi does not teach further comprising: repeating said generating the first higher-resolution image, said generating the residue refinement layer, and said generating second higher-resolution for a plurality of video frames.
However, Xiang teaches further comprising: repeating said generating the first higher-resolution image, said generating the residue refinement layer, and said generating second higher-resolution for a plurality of video frames (Xiang, “Given an LR, LFR video sequence… our goal is to generate the corresponding high-resolution slow-motion video sequence…To fast and accurately increase resolution in both space and time domains, we propose a one-stage space-time super-resolution framework: Zooming Slow-Mo as illustrated in Figure 2. The framework mainly consists of four parts: feature extractor, frame feature temporal interpolation module, deformable ConvLSTM, and HR frame reconstructor.”, pg. 3372, 2nd column, 1st full paragraph).
Esmaeilzehi teaches applying bicubic interpolation to low-resolution images to train a network to predict residual images for image super-resolution (Esmaeilzehi, pg. 3, 1st column, 2nd paragraph, see Figs. 1 and 2). Esmaeilzehi does not teach repeating the steps of claim 7 for a plurality of video frames. Xiang teaches Xiang teaches performing super resolution on sequences of video frames to generate high-resolution videos (see above). Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to have modified the network of Esmaeilzehi to be applied for video frame sequences as taught by Xiang (Xiang, pg. 3372, 2nd column, 1st full paragraph, see Figs. 2 and 3). The motivation for doing so would have been extend Esmaeilzehi’s image super-resolution to video processing, thereby generating visually appealing high-resolution videos (as suggested by Xiang, “Space-Time Video Super-Resolution (STVSR) [30] aims to automatically generate a photo-realistic video sequence with a high space-time resolution from a low-resolution and low frame rate input video. Since HR slow-motion videos are more visually appealing containing fine image details and clear motion dynamics, they are desired in rich applications, such as film making and high-definition television.”, pg. 3370, 2nd column, 1st full paragraph). Further, one skilled in the art could have combined the elements as described above by known methods with no change in their respective functions, and the combination would have yielded nothing more than predictable results. Therefore, it would have been obvious to combine the teachings of Esmaeilzehi with Xiang to obtain the invention according to claim 8.
Claim 17 corresponds to claim 8, with the addition of a non-transitory computer-readable medium or media having stored thereon machine interpretable instructions which, when executed by a processing system, cause the processing system to perform the functions according to claim 1. Esmaeilzehi in view of Xiang teaches the addition of a non-transitory computer-readable medium or media having stored thereon machine interpretable instructions (Esmaeilzehi, “The Keras [38] library and TensorFlow package [39] are used for implementing MuRNet. The training of MuRNet is carried out on the machine with Intel Core i7 CPU @4.2 GHz, 16-GB RAM and GPU Nvidia Titan X (Pascal).”, pg. 5, 1st column, 1st full paragraph). As indicated in the analysis of claim 8, Esmaeilzehi in view of Xiang teaches all the limitation according to claim 8. Therefore, claim 17 is rejected for the same reason of obviousness as claim 8.
Claims 5 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Esmaeilzehi et al. (“MuRNet: A deep recursive network for super resolution of bicubically interpolated images”, Signal Processing: Image Communication, 2021) in view of Ledig et al. (“Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network”, Proceedings of the IEEE conference on computer vision and pattern recognition. 2017.), (hereinafter Ledig).
Regarding claim 5, Esmaeilzehi teaches the image processing system of claim 1. Esmaeilzehi does not teach wherein the neural network is trained as part of a generative adversarial network.
However, Ledig teaches wherein the neural network is trained as part of a generative adversarial network (Ledig, “In this work we propose a super-resolution generative adversarial network (SRGAN) for which we employ a deep residual network (ResNet) with skip-connection and diverge from MSE as the sole optimization target.”, pg. 4682, 1st column, 1st full paragraph, lines 1-4, “The general idea behind this formulation is that it allows one to train a generative model G with the goal of fooling a differentiable discriminator D that is trained to distinguish super-resolved images from real images. With this approach our generator can learn to create solutions that are highly similar to real images and thus difficult to classify by D.”, pg. 4684. 2nd column, 1st full paragraph, lines 6-11).
Esmaeilzehi teaches a training and implementing a convolutional network for single image super-resolution by performing bicubic interpolation on LR images to train the network to predict a residual image (Xiang, pg. 3372, 2nd column, 1st full paragraph, see Figs. 2 and 3). Esmaeilzehi does not teach training this network as part of a Generative Adversarial Network. Ledig teaches performing single image super-resolution by training a Generative Adversarial Network that includes a deep residual network generator (see above). Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to have modified the network architecture of Esmaeilzehi to be a generative adversarial network as taught by Ledig (Ledig, pg. 4682, 1st column, 1st full paragraph, lines 1-4 and pg. 4684. 2nd column, 1st full paragraph, lines 6-11). The motivation for doing so would have been generate more photo-realistic and perceptually convincing images by leveraging adversarial loss (as suggested by Ledig, “GANs provide a powerful framework for generating plausible-looking natural images with high perceptual quality. The GAN procedure encourages the reconstructions to move towards regions of the search space with high probability of containing photo-realistic images and thus closer to the natural image manifold as shown in Figure 3.”, pg. 4684, 1st column, 1st full paragraph). Further, one skilled in the art could have combined the elements as described above by known methods with no change in their respective functions, and the combination would have yielded nothing more than predictable results. Therefore, it would have been obvious to combine the teachings of Esmaeilzehi with Ledig to obtain the invention according to claim 5.
Claim 15 corresponds to claim 5, reciting a computer-implemented method to perform the steps according to claim 5. Esmaeilzehi in view of Ledig teaches a computer-implemented method to perform the steps according to claim 5 (Esmaeilzehi, “The Keras [38] library and TensorFlow package [39] are used for implementing MuRNet. The training of MuRNet is carried out on the machine with Intel Core i7 CPU @4.2 GHz, 16-GB RAM and GPU Nvidia Titan X (Pascal).”, pg. 5, 1st column, 1st full paragraph, see Figs. 1 and 2). As indicated in the analysis of claim 5, Esmaeilzehi in view of Ledig teaches all of the limitations according to claim 5. Therefore, claim 15 is rejected for the same reason as claim 5.
Claims 10 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Esmaeilzehi et al. (“MuRNet: A deep recursive network for super resolution of bicubically interpolated images”, Signal Processing: Image Communication, 2021) in view of Olsen et al. (US 20240211540 A1), (hereinafter Olsen).
Regarding claim 10, Esmaeilzehi teaches the computer-implemented method of claim 7. Esmaeilzehi does not teach further comprising transmitting the second higher-resolution image to a client device.
However, Olsen teaches further comprising transmitting the second higher-resolution image to a client device (Olsen, “At operation 824, one or more of the predicted high-resolution images are stored, transmitted, displayed, or further processed. For example, the third predicted high-resolution image may be transmitted to one or more client devices for display or processing of the image. Current high-resolution images may have many uses and applications, including managing or monitoring agricultural growth and development, forest management, or any other application where detailed images may be beneficial.”, pg. 8, paragraph 0075).
Esmaeilzehi teaches training and implementing a single image super-resolution network to produce high-resolution images (Esmaeilzehi, see Fig. 1, HR Image). Esmaeilzehi does not teach transmitting output high-resolution images to a client device. Olsen teaches predicted high-resolution images and transmitting them to a client device for further display or processing (see above). Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to have modified the network of Esmaeilzehi to transmit the output high-resolution images to a client device as taught by Olsen (Olsen, pg. 8, paragraph 0075). The motivation for doing so would have been to provide high quality images directly to a client for display or processing. Further, one skilled in the art could have combined the elements as described above by known methods with no change in their respective functions, and the combination would have yielded nothing more than predictable results. Therefore, it would have been obvious to combine the teachings of Esmaeilzehi with Olsen to obtain the invention according to claim 10.
Claim 19 corresponds to claim 10, with the addition of a non-transitory computer-readable medium or media having stored thereon machine interpretable instructions which, when executed by a processing system, cause the processing system to perform the functions according to claim 10. Esmaeilzehi in view of Olsen teaches the addition of a non-transitory computer-readable medium or media having stored thereon machine interpretable instructions (Esmaeilzehi, “The Keras [38] library and TensorFlow package [39] are used for implementing MuRNet. The training of MuRNet is carried out on the machine with Intel Core i7 CPU @4.2 GHz, 16-GB RAM and GPU Nvidia Titan X (Pascal).”, pg. 5, 1st column, 1st full paragraph). As indicated in the analysis of claim 10, Esmaeilzehi in view of Olsen teaches all the limitation according to claim 10. Therefore, claim 19 is rejected for the same reason of obviousness as claim 10.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CONNOR LEVI HANSEN whose telephone number is (703)756-5533. The examiner can normally be reached Monday-Friday 9:00-5:00 (ET).
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, Sumati Lefkowitz can be reached at (571) 272-3638. 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.
/CONNOR L HANSEN/Examiner, Art Unit 2672
/SUMATI LEFKOWITZ/Supervisory Patent Examiner, Art Unit 2672
Prosecution Insights