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 .
Priority
Receipt is acknowledged that application is a National Stage application of PCT/US2021/054721. Priority to PCT/US2021/054721 with a priority date of 10/13/2021 is acknowledged under 35 USC 119(e) and 37 CFR 1.78.
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's RCE submission filed on 08/06/2026 has been entered.
Response to Amendment
The amendment filed 08/06/2026 has been entered. Claims 1-2, 4-6, 8-10, 12-13, 15-18, and 20-25 remain pending in the application, with claims 7, 14, and 19 being newly cancelled and claims 23-25 being newly added.
Response to Arguments
Applicant’s arguments have been considered but are moot because the new ground of rejection does not rely on any combination of references applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
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 1, 2, 4, 6, and 23-25 are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al. (CN Patent No. 113240581 A), hereinafter Wang, in view of Wang et al. (Wang, Y., Yang, J., Wang, L., Ying, X., Wu, T., An, W., & Guo, Y. (2020). Light field image super-resolution using deformable convolution. IEEE Transactions on Image Processing, 30, 1057-1071), hereinafter Guo.
Regarding claim 1, Wang teaches a computing device, comprising: a processor resource; and a non-transitory memory resource storing machine-readable instructions (Wang, para n0042: “i9-8700k processor, 16GB of RAM, an NVIDIA GeForce GTX1080Ti 8GB graphics card, and a Windows operating system”) that, when executed, cause the processor resource to:
identify a base resolution of a captured image having a base image quality (Wang, resolution and quality of low-resolution images input to the model, para n0013: “low-resolution image LR for training the network”; see low-resolution image, LR, in FIG. 1, top image attached below); and
perform, via an individual neural network (Wang, super-resolution network in Figure 1, para n0023, para n0038), a plurality of neural network calculations on the captured image (Wang, steps outlined in para n0038) to form an enhanced image (Wang, para n0014: “output a high-resolution RGB image”; see high-resolution image, SR, in FIG. 1) having:
an increased resolution that is higher than the base resolution (Wang, para n0017: “This invention proposes a real-world image super-resolution network for images with unknown fuzzy kernels, which is applicable to amplifying and improving the resolution of real-world images”); and
an increased image quality that is higher than the base image quality, wherein the individual neural network is to form the enhanced image having the increased image quality by reduction of noise in the captured image (Wang, para n0052: “Since the degradation process involves blurring and noise injection, the signal power and noise power are significantly improved”; para n0010: “The resulting noise-free image is considered the high-resolution image”; see RRDB-SFT results in Figure 3, bottom image attached below).
PNG
media_image1.png
285
1000
media_image1.png
Greyscale
Figure 1
PNG
media_image2.png
572
1000
media_image2.png
Greyscale
Figure 3
Wang teaches wherein the individual neural network comprises a convolution layer (Wang, first layer after the LR image in FIG. 1), a residual-in-residual dense block (RRDB) (Wang, para n0038: “The second part is the nonlinear mapping base block, which adopts the Dense Block and is connected in an RRDB structure”), and a subpixel convolution layer (Wang, para n0038: “The third part uses subpixel convolution to amplify the feature map and then convolutions to generate an RGB three-channel image”). However, Wang fails to explicitly teach a pooling layer, and therefore fails to teach: an atrous spatial pyramid pooling (ASPP) layer and wherein the ASPP layer is a preceding layer to the RRDB in the individual neural network.
However, Guo similarly teaches a method for image super-resolution using an individual neural network comprising a convolution layer, residual blocks, and a pixel shuffle layer (Guo, see following citations and FIG. 1 on pg. 4). Within the model’s architecture, Guo teaches a method for feature extraction, including an atrous spatial pyramid pooling (ASPP) layer (Guo, pg. 3, section A: “Discriminative feature representation with rich spatial context information is beneficial to the subsequent feature alignment and SR reconstruction steps. Therefore, a large receptive field with a dense pixel sampling rate is required to extract hierarchical features. To this end, we follow [61] and use residual atrous spatial pyramid pooling (ASPP) module as the feature extraction module in our LF-DFnet”) and wherein the ASPP layer is a preceding layer to a residual block in the individual neural network (Guo, pg. 3, section A: “As shown in Fig. 1(a), input SAIs are first processed by a 1 ×1 convolution to generate initial features, and then fed to residual ASPP modules (Fig. 1(b)) and residual blocks (Fig. 1(c)) for deep feature extraction”; see excerpt from FIG. 1 below).
PNG
media_image3.png
207
277
media_image3.png
Greyscale
From Figure 1 of Guo
It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the ASPP layer before the residual blocks, taught by Guo above, with the RRDB blocks in the neural network taught by Wang in order to improve the feature extraction of the RRDB blocks by first extracting multi-scale features with rich spatial context information (Guo, pg. 11, section C2: “That is because, residual ASPP module can extract hierarchical features from input images, which are beneficial to LF image SR…That is because, the ASPP module can achieve accurate offset learning through multi scale feature representation and the enlargement of receptive fields”). The same benefit that preceding ASPP layers provide to the residual blocks of Guo would apply to the RRDB blocks of Wang. Implementing an ASPP layer before the RRDB blocks would similarly improve the super-resolution by allowing them to retain more input image features/details using the hierarchical, multi-scale contextual information received from the ASPP layer, thus improving the reconstruction of image details in the super-resolution image.
Regarding claim 2 (dependent on claim 1), Wang in view of Guo teaches wherein the processor resource is to perform the plurality of neural network calculations during inference (Wang, see para n0042-n0052 describing experiments using the models).
Regarding claim 4 (dependent on claim 1), Wang in view of Guo teaches wherein the neural network is a convolutional neural network (CNN) and wherein the plurality of neural network calculations are a plurality of CNN calculations (Wang, convolutional calculations described in para n0038).
Regarding claim 6 (dependent on claim 1), Wang in view of Guo teaches wherein the increased resolution is 1.5 times or greater than the base resolution (Wang, see RRDB-SFT results in the right column of Figure 3, attached with claim 1 – pixel resolution for the boxed area is increased by at least 1.5 times).
Regarding claim 23 (dependent on claim 1), Wang in view of Guo teaches wherein the individual neural network comprises an ordered sequence of layers, and wherein in the ordered sequence of layers: the convolution layer precedes the ASPP layer, the ASPP layer precedes the RRDB, and the RRDB precedes the subpixel convolution layer (Taught by Wang in view of Guo in the claim 1 rejection – Wang teaches the ordered sequence of convolution layer, RRDB, and subpixel layer; see FIG. 3 of Wang. Guo teaches the addition of an ASPP layer preceding the RRDB; see FIG. 1 of Guo.).
Regarding claim 24 (dependent on claim 23), Wang in view of Guo teaches wherein the individual neural network comprises an upsampling connection to upsample the captured image, and wherein the individual neural network is to combine the upsampled image and an output of the subpixel convolution layer to form the enhanced image (Guo, see combining of the Bicubic upscaling with the output from the Pixel Shuffle in FIG. 1, excerpt attached below).
Wang in view of Guo teaches a neural network for generating an enhanced image with increased resolution, but fails to explicitly teach the upsampling steps recited in claim 24; however, Guo teaches the combining of two upscaled image data sets to form the enhanced image. One of ordinary skill in the art, before the effective filing date of the claimed invention, could have combined the steps, outlined above, taught by Guo, with the neural network taught by Wang in view of Guo, using known methods. In doing so, each element merely would have performed the same functions as it did separately (image super-resolution) and would achieve the predictable results of generating an enhanced image with increased resolution. Additionally, utilizing this method can increase the capacity of the neural network model and its efficiency by reducing the amount of data that flows through the convolutional layers of the neural network model.
PNG
media_image4.png
210
903
media_image4.png
Greyscale
From Figure 1 of Guo
Regarding claim 25 (dependent on claim 1), Wang in view of Guo teaches wherein image data associated with the captured image sequentially propagates through the convolution layer, the ASPP layer, the RRDB, and the subpixel convolution layer for generation of the enhanced image (Taught by Wang in view of Guo in the claim 1 rejection – Wang teaches the ordered sequence of convolution layer, RRDB, and subpixel layer; see FIG. 3 of Wang. Guo teaches the addition of an ASPP layer preceding the RRDB; see FIG. 1 of Guo.).
Claims 5, 8, 13, and 17-18 are rejected under 35 U.S.C. 103 as being unpatentable over Wang in view of Guo, in further view of Yang et al. (U.S. Patent No. 2021/0287342 A1), hereinafter Yang.
Regarding claim 5 (dependent on claim 4), Wang in view of Guo teaches wherein the CNN calculations include CNN calculations to:
segment, via the convolution layer of the CNN, the captured image into image
dimensions having a given height and a given width (Wang, first convolutional layer in Figure 1, para n0041: “first convolution”);
extract features from the image dimensions (Wang, para n0038: “The second part is the nonlinear mapping base block, which adopts the Dense Block and is connected in an RRDB structure”); and
perform subpixel convolution based on the extracted features and the image dimensions to form the enhanced image (Wang, para n0038: “The third part uses subpixel convolution to amplify the feature map and then convolutions to generate an RGB three-channel image”).
While Wang in view of Guo teaches the ASPP layer (see claim 1 rejection), Wang in view of Guo fails to explicitly teach pool, via the ASPP layer of the CNN, the image dimensions to form pooled image dimensions (emphasis added).
However, Yang teaches a convolutional neural network for increasing image quality (see abstract), further disclosing: pool, via an ASPP layer of the CNN, image dimensions to form pooled image dimensions (Yang, para 49: “The image pooling layer 328 computes global features based on the input features (e.g., computes a global average or a global maximum over all of the features in the input to the ASPP 320)”). Features are then extracted from the pooled image dimensions using residual dense blocks (Yang, para 55: “feature maps from earlier modules via the residual connections to compute an intermediate feature map, which is concatenated with the output of the concatenation module 330 of the ASPP module 320 and compresses the concatenated result using a cony 1×1 layer. The output of the concatenation module 358 is added to the input feature map 302 by an adder 360 to compute an output feature map 392 of the MRDB”).
In the same way as Wang in view of Guo, the residual dense blocks of Yang utilize atrous spatial pyramid pooling to compute multi-scale features for a deeper feature extraction. Wang in view of Guo discloses a base method for utilizing an ASPP layer, but does not specify specific methods for forming pooled image dimensions. Yang teaches a known technique of utilizing an ASPP layer to form pooled image dimensions. A person having ordinary skill in the art, before the effective filing date of the claimed invention, could have applied the known technique, as taught by Yang, in the same way to the computing device of Wang in view of Guo and achieved predictable results of reducing image dimensions during neural network processing.
Implementing pooling after the first convolutional layer of Wang results in subsequent processing steps being performed on the pooled image dimensions, such as extract features from the pooled image dimensions and perform subpixel convolution based on the pooled image dimensions.
Regarding claim 8, Wang teaches a non-transitory memory resource storing machine-readable instructions (Wang, para n0042: “i9-8700k processor, 16GB of RAM, an NVIDIA GeForce GTX1080Ti 8GB graphics card, and a Windows operating system”) that, when executed, cause a processor resource to:
identify a base resolution of a captured image having a base image quality (Wang, resolution and quality of low-resolution images input to the model, para n0013: “low-resolution image LR for training the network”; see low-resolution image, LR, in FIG. 1); and
perform, via an individual convolutional neural network (CNN) (Wang, super-resolution network in Figure 1, para n0023, convolutional calculations described in para n0038), a plurality of CNN calculations on the captured image to form an enhanced image having an increased resolution that is higher than the base resolution (Wang, para n0017: “This invention proposes a real-world image super-resolution network for images with unknown fuzzy kernels, which is applicable to amplifying and improving the resolution of real-world images”) and an increased image quality that is higher than the base image quality (Wang, increase quality by removing noise from input images, para n0052: “Since the degradation process involves blurring and noise injection, the signal power and noise power are significantly improved”; para n0010: “The resulting noise-free image is considered the high-resolution image”; see RRDB-SFT results in Figure 3), the CNN calculations including CNN calculations to:
segment the captured image into image dimensions having a given height and a given width (Wang, first convolutional layer in Figure 1, para n0041: “first convolution”);
extract features from the image dimensions using a residual-in-residual dense block (RRDB) (Wang, para n0038: “The second part is the nonlinear mapping base block, which adopts the Dense Block and is connected in an RRDB structure”); and
perform subpixel convolution based on the extracted features and the image dimensions to form the enhanced image (Wang, para n0038: “The third part uses subpixel convolution to amplify the feature map and then convolutions to generate an RGB three-channel image”).
Wang fails to explicitly teach a pooling layer, and therefore fails to teach: pool, via an atrous spatial pyramid pooling (ASPP) layer of the CNN, the image dimensions to form pooled image dimensions; and wherein the ASPP layer of the CNN is a preceding layer to the RRDB in the CNN.
However, Guo similarly teaches a method for image super-resolution using an individual neural network comprising a convolution layer, residual blocks, and a pixel shuffle layer (Guo, see following citations and FIG. 1 on pg. 4). Within the model’s architecture, Guo teaches a method for feature extraction, including an atrous spatial pyramid pooling (ASPP) layer of the CNN (Guo, pg. 3, section A: “Discriminative feature representation with rich spatial con text information is beneficial to the subsequent feature alignment and SR reconstruction steps. Therefore, a large receptive field with a dense pixel sampling rate is required to extract hierarchical features. To this end, we follow [61] and use residual atrous spatial pyramid pooling (ASPP) module as the feature extraction module in our LF-DFnet”) and wherein the ASPP layer of the CNN is a preceding layer to a residual block in the CNN (Guo, pg. 3, section A: “As shown in Fig. 1(a), input SAIs are first processed by a 1 ×1 convolution to generate initial features, and then fed to residual ASPP modules (Fig. 1(b)) and residual blocks (Fig. 1(c)) for deep feature extraction”; see FIG. 1 on pg. 4).
It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the ASPP layer before the residual blocks, taught by Guo above, with the RRDB blocks in the neural network taught by Wang in order to improve the feature extraction of the RRDB blocks by first extracting multi-scale features with rich spatial context information (Guo, pg. 11, section C2: “That is because, residual ASPP module can extract hierarchical features from input images, which are beneficial to LF image SR…That is because, the ASPP module can achieve accurate offset learning through multi scale feature representation and the enlargement of receptive fields”). The same benefit that preceding ASPP layers provide to the residual blocks of Guo would apply to the RRDB blocks of Wang. Implementing an ASPP layer before the RRDB blocks would similarly improve the super-resolution by allowing them to retain more input image features/details using the hierarchical, multi-scale contextual information received from the ASPP layer, thus improving the reconstruction of image details in the super-resolution image.
Furthermore, while Wang in view of Guo teaches the ASPP layer, Wang in view of Guo fails to explicitly teach pool, via the ASPP layer of the CNN, the image dimensions to form pooled image dimensions (emphasis added).
However, Yang teaches a convolutional neural network for increasing image quality (see abstract), further disclosing pool, via an ASPP layer of the CNN, image dimensions to form pooled image dimensions (Yang, para 49: “The image pooling layer 328 computes global features based on the input features (e.g., computes a global average or a global maximum over all of the features in the input to the ASPP 320)”). Features are then extracted from the pooled image dimensions using residual dense components (Yang, para 55: “feature maps from earlier modules via the residual connections to compute an intermediate feature map, which is concatenated with the output of the concatenation module 330 of the ASPP module 320 and compresses the concatenated result using a cony 1×1 layer. The output of the concatenation module 358 is added to the input feature map 302 by an adder 360 to compute an output feature map 392 of the MRDB”).
In the same way as Wang in view of Guo, the residual dense blocks of Yang utilize atrous spatial pyramid pooling to compute multi-scale features for a deeper feature extraction. Wang in view of Guo discloses a base method for utilizing an ASPP layer, but does not specify specific methods for forming pooled image dimensions. Yang teaches a known technique of utilizing an ASPP layer to form pooled image dimensions. A person having ordinary skill in the art, before the effective filing date of the claimed invention, could have applied the known technique, as taught by Yang, in the same way to the computing device of Wang in view of Guo and achieved predictable results of reducing image dimensions during neural network processing.
Implementing pooling after the first convolutional layer of Wang results in subsequent processing steps being performed on the pooled image dimensions, such as extract features from the pooled image dimensions and perform subpixel convolution based on the pooled image dimensions.
Regarding claim 13, Wang teaches a computing device (Wang, para n0042: “i9-8700k processor, 16GB of RAM, an NVIDIA GeForce GTX1080Ti 8GB graphics card, and a Windows operating system”), comprising: an image capture device to capture an image with a base resolution and a base image quality (Wang, para n0042: “The DPED dataset contains 5,614 images taken by the iPhone 3 camera. This dataset consists of unprocessed real-world images, including low-quality issues such as noise and blur”). All further claim limitations are met and rendered obvious by Wang in view of Guo and Yang because the steps performed by the processor resource in claim 13 are the same as those recited in claim 8.
Regarding claim 17 (dependent on claim 5), Wang in view of Guo and Yang teaches wherein the processor resource is to extract the features from the pooled image dimensions using the RRDB (Wang, para n0038: “The second part is the nonlinear mapping base block, which adopts the Dense Block and is connected in an RRDB structure”; pooled image dimensions taught in combination with Yang in the rejection of claim 5).
Regarding claim 18 (dependent on claim 5), Wang in view of Guo and Yang teaches wherein forming the enhanced image comprises: generating first upscaled image data based on upsampling the captured image; generating second upscaled image data based on performing the subpixel convolution based on the extracted features and the pooled image dimensions; and combining the first upscaled image data and the second upscaled image data to form the enhanced image (Guo, see combining of the Bicubic upscaling with the output from the Pixel Shuffle in FIG. 1).
Wang in view of Guo teaches a neural network for generating an enhanced image with increased resolution, but fails to explicitly teach the upsampling steps recited in claim 18; however, Guo teaches the combining of two upscaled image data sets to form the enhanced image. One of ordinary skill in the art, before the effective filing date of the claimed invention, could have combined the steps, outlined above, taught by Guo, with the neural network taught by Wang in view of Guo and Yang, using known methods. In doing so, each element merely would have performed the same functions as it did separately (image super-resolution) and would achieve the predictable results of generating an enhanced image with increased resolution. Additionally, utilizing this method can increase the capacity of the neural network model and its efficiency by reducing the amount of data that flows through the convolutional layers of the neural network model.
Claims 9-10, 15, and 20-22 are rejected under 35 U.S.C. 103 as being unpatentable over Wang in view of Guo, in further view of Yang and Caballero et al. (cited in IDS: U.S. Patent No. 10,701,394 B1), hereinafter Caballero.
Regarding claim 9 (dependent on claim 8), Wang in view of Guo and Yang fails to explicitly teach wherein the captured image has undergone lossy image compression to reduce an image resolution of the captured image from an original resolution to the base resolution.
However, Caballero teaches an image processing method wherein the captured image has undergone lossy image compression to reduce an image resolution of the captured image from an original resolution to the base resolution (Caballero, col 44, ln 38-40: “The compression of the scene can be implemented using any well-known lossy image/video compression algorithm”). Wang in view of Yang discloses a base method for increasing the resolution and quality of low-resolution images, but does not specify specific methods for how the image was compressed from its original resolution. Caballero teaches the known technique of a lossy image compression method. A person having ordinary skill in the art, before the effective filing date of the claimed invention, could have applied the known technique, as taught by Caballero, in the same way to the method performed by the device of Wang in view of Guo and Yang and achieved predictable results of utilizing a well-known image compression technique to reduce file size and preserve memory resources.
Regarding claim 10 (dependent on claim 9), Wang in view of Guo and Yang teaches wherein the individual CNN is to form the enhanced image having the increased image quality by: sharpening feature boundaries dulled by the lossy image compression (Wang, removing noise from an image also sharpens features – see sharper feature boundaries in SR image versus input LR in Figure 1 and in Figure 3; lossy image compression taught in combination with Caballero in the rejection of claim 9).
Regarding claim 15 (dependent on claim 13), Wang in view of Guo and Yang fails to explicitly teach wherein the computing device further comprises a first computing device that is communicatively coupled to and in a teleconference with a second computing device, and wherein the processor resource is to: receive, during the teleconference, the image data of the image from an image capture device of the first computing device; and provide the enhanced image to the second computing device during the teleconference.
However, Caballero teaches an image processing method for increasing image resolution (see Figure 16 below) including: a first computing device that is communicatively coupled to and in a teleconference with a second computing device (see citations below), and wherein the processor resource is to: receive, during the teleconference, the image data of the image from an image capture device of the first computing device (Caballero, col 36, ln 25-40: “a technique 1500 is used to increase the resolution of visual data will now be described in detail. These embodiments can be used in combination with other embodiments described elsewhere in this specification. Received video data 1540 is provided into a decoder system and is a lower-resolution video encoded in a standard video format… The system then separates video data 1510 into single frames at step 1520, i.e. into a sequence of images at the full resolution of the received video data 310”); and provide the enhanced image to the second computing device during the teleconference (Caballero, col 38, ln 7-20: “Further, output video 1570 in process 1500 or 1600 can be output directly to a display or may be stored for viewing on a display on a local or remote storage device, or forwarded to a remote node for storage or viewing as required. The input video concerned may be media for playback, such as recorded video or live streamed video, or it can be videoconference video or any other video source such as video recorded or being recorded on a portable device such as a mobile phone or a video recording device such as a video camera or surveillance camera”).
PNG
media_image5.png
330
612
media_image5.png
Greyscale
It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the communications between devices, as taught by Caballero above, with the system of Wang in view of Guo and Yang in order to output images with improved quality and resolution to improve their use in real-world applications, such as teleconference (See col 38 citation from Caballero above).
Regarding claim 20 (dependent on claim 9), Wang in view of Guo and Yang teaches wherein the individual CNN is to form the enhanced image having the increased image quality by: reduction of noise imparted by the lossy image compression (See claim 8 rejection wherein the method of Wang reduces noise from the image; lossy image compression taught by Caballero in claim 9 rejection).
Regarding claim 21 (dependent on claim 9), Wang in view of Guo and Yang teaches wherein the individual CNN is to form the enhanced image having the increased image quality by: reduction of compression artifacts imparted by the lossy image compression (Reducing noise and increasing resolution, taught by Wang in claim 8 rejection, reduces compression artifacts, which can include noise; lossy image compression taught by Caballero in claim 9 rejection).
The method of Wang in view of Guo and Yang reduces noise in the low-resolution image (See claim 1 rejection). Accordingly, in combination with the compressed image data taught by Caballero, the invention of Wang in view of Guo, Yang, and Caballero reduces artifacts imparted by compression by increasing resolution and removing noise (Compression artifacts include noise; see image improvements in the claim 1 rejection, demonstrated by Figures 1 and 3 of Wang). Therefore, Wang in view of Guo, Yang, and Caballero teaches “reduction of compression artifacts imparted by the lossy image compression.”
Regarding claim 22 (dependent on claim 13), Wang in view of Guo and Yang fails to explicitly teach wherein the captured image has undergone lossy image compression, and wherein the individual CNN is to form the enhanced image having the increased image quality by: reduction of compression artifacts imparted by the lossy image compression.
However, Caballero teaches an image processing method wherein the captured image has undergone lossy image compression to reduce an image resolution of the captured image from an original resolution to the base resolution (Caballero, col 44, ln 38-40: “The compression of the scene can be implemented using any well-known lossy image/video compression algorithm”). Wang in view of Guo and Yang discloses a base method for increasing the resolution and quality of low-resolution images, but does not specify specific methods for how the image was compressed from its original resolution. Caballero teaches the known technique of a lossy image compression method. A person having ordinary skill in the art, before the effective filing date of the claimed invention, could have applied the known technique, as taught by Caballero, in the same way to the method performed by the device of Wang in view of Guo and Yang and achieved predictable results of utilizing a well-known image compression technique to reduce file size and preserve memory resources.
The method of Wang in view of Guo and Yang reduces noise in the low-resolution image (See claim 1 rejection). Accordingly, in combination with the compressed image data taught by Caballero, the invention of Wang in view of Guo, Yang, and Caballero reduces artifacts imparted by compression by increasing resolution and removing noise (Compression artifacts include noise; see image improvements in the claim 1 rejection, demonstrated by Figures 1 and 3 of Wang). Therefore, Wang in view of Guo, Yang, and Caballero teaches “reduction of compression artifacts imparted by the lossy image compression.”
Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Wang in view of Guo, in further view of Yang and El-Khamy et al. (U.S. Patent No. 2020/0090305 A1), hereinafter El-Khamy.
Regarding claim 12 (dependent on claim 8), Wang in view of Guo and Yang fails to explicitly teach wherein the processor resource is to determine: an upsampling factor; and utilize the upsampling factor to form the enhanced image.
However, El-Khamy teaches a super resolution method (See abstract), including an upsampling factor; and utilize the upsampling factor to form the enhanced image (El-Khamy, para 59: “the first individual super resolution network S.sub.1 receives the LR input image 202, or alternatively, a bicubic upsampled version of the LR input image 202, where the upsampling ratio is according to a target upsampling ratio”). It would have been obvious to a person having ordinary skill in the art, before the effective filing date of the claimed invention, to have combined the upsampling factor of El-Khamy with the device of Wang in view of Guo and Yang in order to control the spatial resolution of the inputs/outputs of the model (El-Khamy, see para 59 and equation).
Claim 16 is rejected under 35 U.S.C. 103 as being unpatentable over Wang in view of Guo, in further view of Caballero.
Regarding claim 16 (dependent on claim 1), Wang in view of Guo fails to explicitly teach wherein the captured image has undergone lossy image compression, and wherein the individual neural network is to form the enhanced image having the increased image quality by: reduction of compression artifacts imparted by the lossy image compression.
However, Caballero teaches an image processing method wherein the captured image has undergone lossy image compression to reduce an image resolution of the captured image from an original resolution to the base resolution (Caballero, col 44, ln 38-40: “The compression of the scene can be implemented using any well-known lossy image/video compression algorithm”). Wang in view of Guo discloses a base method for increasing the resolution and quality of low-resolution images, but does not specify specific methods for how the image was compressed from its original resolution. Caballero teaches the known technique of a lossy image compression method. A person having ordinary skill in the art, before the effective filing date of the claimed invention, could have applied the known technique, as taught by Caballero, in the same way to the method performed by the device of Wang in view of Guo and achieved predictable results of utilizing a well-known image compression technique to reduce file size and preserve memory resources.
The method of Wang in view of Guo reduces noise in the low-resolution image (See claim 1 rejection). Accordingly, in combination with the compressed image data taught by Caballero, the invention of Wang in view of Guo and Caballero reduces artifacts imparted by compression by increasing resolution and removing noise (Compression artifacts include noise; see image improvements in the claim 1 rejection, demonstrated by Figures 1 and 3 of Wang). Therefore, Wang in view of Guo and Caballero teaches “reduction of compression artifacts imparted by the lossy image compression.”
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Wen (CN Patent No. 112070669 A) teaches a similar convolutional super-resolution model.
PNG
media_image6.png
338
710
media_image6.png
Greyscale
Wang et al. (Wang, H., Su, D., Liu, C., Jin, L., Sun, X., & Peng, X. (2019). Deformable non-local network for video super-resolution. IEEE Access, 7, 177734-177744.) teaches a super-resolution model using residual in residual dense blocks.
PNG
media_image7.png
283
1014
media_image7.png
Greyscale
Shang et al. (Shang, T., Dai, Q., Zhu, S., Yang, T., & Guo, Y. (2020, June). Perceptual extreme super resolution network with receptive field block. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) (pp. 1778-1787). IEEE.) teaches a convolutional super-resolution model using residual in residual dense blocks.
PNG
media_image8.png
624
691
media_image8.png
Greyscale
Lijun et al. (Lijun, Y., Xiaoming, Z., Fan, L., Gang, S., Zhou, C., Jing, Y., ... & Qiang, T. (2021, March). Image super-resolution method based on generative adversarial network. In 2021 4th International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE) (pp. 909-915). IEEE.) teaches a super-resolution model using spatial pyramid pooing and residual in residual dense blocks.
Wang et al. (Wang, X., Yu, K., Wu, S., Gu, J., Liu, Y., Dong, C., ... & Loy, C. C. (2018, September). Esrgan: Enhanced super-resolution generative adversarial networks. In European conference on computer vision (pp. 63-79). Cham: Springer International Publishing.) teaches a convolutional super-resolution model using residual in residual dense blocks.
PNG
media_image9.png
815
752
media_image9.png
Greyscale
Zhang et al. (Zhang, Y., Li, K., Li, K., Wang, L., Zhong, B., & Fu, Y. (2018, September). Image super-resolution using very deep residual channel attention networks. In European conference on computer vision (pp. 294-310). Cham: Springer International Publishing.) teaches a convolutional super-resolution model using residual in residual dense blocks.
PNG
media_image10.png
239
608
media_image10.png
Greyscale
Bao, L., Yang, Z., Wang, S., Bai, D., & Lee, J. (2020, June). Real image denoising based on multi-scale residual dense block and cascaded U-Net with block-connection. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) (pp. 1823-1831). IEEE.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to EMMA E DRYDEN whose telephone number is (571)272-1179. The examiner can normally be reached M-F 9-5 EST.
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, ANDREW BEE can be reached at (571) 270-5183. 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.
/EMMA E DRYDEN/Examiner, Art Unit 2677
/ANDREW W BEE/Supervisory Patent Examiner, Art Unit 2677
Prosecution Insights