DETAILED ACTION
Response to Amendment
This Action is responsive to Applicant’s response filed on 07/29/2026. All claims are still pending
in the present application. This Action is made FINAL.
Amendment
Applicants submitted amendments on 07/29/2026. The Examiner acknowledges the amendment and has reviewed the claims accordingly.
Response to Arguments
During prosecution, claim scope not solely on the basis of claim language, but also on giving claims their broadest reasonable construction in light of the specification as it would be interpreted by one of ordinary skill in the art. In re Am. Acad. of Sci. Tech. Ctr., 367 F.3d 1359, 1364 (Fed. Cir. 2004). See also Superguide Corp. v. DirecTV Enterprises, Inc., 358 F.3d 870, 875 (Fed. Cir. 2004) (“Though understanding the claim language may be aided by explanations contained in the written description, it is important not to import into a claim limitations that are not part of the claim.”). Additionally, “[t]hough understanding the claim language may be aided by the explanations contained in the written description, it is important not to import into a claim limitations that are not a part of the claim. For example, a particular embodiment appearing in the written description may not be read into a claim when the claim language is broader than the embodiment.” Superguide Corp. v. DirecTV Enterprises, Inc., 358 F.3d 870, 875 (Fed. Cir. 2004).
Applicant argues that Li does not teach/suggest “generating a low-resolution feature quantity of a lower resolution from the high-resolution feature quantity,” because Li's HR reference image and LR input image are processed through separate encoder streams, such that the features generated from the LR input image are not generated from the features of the HR reference image.
Applicant’s arguments have been fully considered but are not persuasive. The Examiner respectfully disagrees. Although Li discloses separate HR and LR image inputs in the embodiment relied upon by Applicant, Li further expressly teaches within each encoder a sequential feature-generation process in which a higher-resolution feature quantity is generated and subsequently downscaled to generate a lower-resolution feature quantity. Specifically, Li teaches that “the encoders (506/508) may consist of a set of initial convolution layers with a stride of 1 to extract low-level features from the input region.” Li then teaches that “The following three subsequent convolution layers with a stride of 2 aim to transform the low-level features to more abstract latent space by inflating the number of channels and enlarging the receptive field via downscaling them to feature maps, e.g., of size (H/8) x (W/8) x (Cx8)” (Li, Paragraph [0050]). Thus, Li's initial stride-1 convolution layers generate feature quantities without spatial downscaling, and the subsequent stride-2 convolution layers operate on those previously generated feature quantities and downscale them to feature maps having a lower spatial resolution. Accordingly, the lower-resolution feature quantity is generated from the previously generated higher-resolution feature quantity, as presently claimed. Applicant's argument concerning the separate HR reference image and LR input image encoder streams therefore does not address this additional teaching of Li within each individual encoder.
As a result, the argued features are written such that they read upon the cited references. Therefore, the previous rejection still applies.
In regards to Argument(s), Applicant(s) state(s) that, Li et al (US 2023/0098437 A1) in view of Mo et al (“Frequency Attention Network: Blind Noise Removal for Real Images”) does not disclose/teach/suggest on the amended claim(s) “execute attention processing only on the low-resolution feature quantity”, therefore, the rejection of 35 U.S.C. 103 should be removed, (Emphasis added, Remarks, page 7-9).
Applicant’s arguments have been considered but are moot in view of the new ground(s) of rejection in view of Li et al (US 2023/0098437 A1) and Srinivas et al (Bottleneck Transformers for Visual Recognition).
Office Action Summary
Claim(s) 6-15 is/are withdrawn.
Claim(s) 1-2, 4, 16-18, and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Li et al (US 2023/0098437 A1) in view of Srinivas et al (Bottleneck Transformers for Visual Recognition).
Claim(s) 3, 5, and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Li et al (US 2023/0098437 A1) in view of Srinivas et al (Bottleneck Transformers for Visual Recognition), further in view of Mo et al (“Frequency Attention Network: Blind Noise Removal for Real Images”).
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) 1-2, 4, 16-18, and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Li et al (US 2023/0098437 A1) in view of Srinivas et al (Bottleneck Transformers for Visual Recognition).
Regarding claim 1, Li teaches an information processing apparatus, comprising:
at least one processor (Figure 8; and Paragraph [0076]); and
a memory coupled to the at least one processor, the memory storing instructions that, when executed by the at least one processor (Figure 8; and Paragraph [0076]), cause the at least one processor to:
generate at least one high-resolution feature quantity for an input image (Figure 5; and Paragraph [0050]: “[…] HR reference image region 502 […] e.g., each with a size of HxWx3, two distinct encoders (506/508) firstly embed them into intermediate feature maps”);
generate a low-resolution feature quantity of a lower resolution from the high-resolution feature quantity (Figure 5; and Paragraph [0050]: “Given an LR input image region 504 […] e.g., each with a size of HxWx3, two distinct encoders (506/508) firstly embed them into intermediate feature maps […] the encoders (506/508) may consist of a set of initial convolution layers with a stride of 1 to extract low-level features from the input region. The following three subsequent convolution layers with a stride of 2 aim to transform the low-level features to more abstract latent space by inflating the number of channels and enlarging the receptive field via downscaling them to feature maps, e.g., of size (H/8) × (W/8) × (Cx8)”); and
combine the high-resolution feature quantity and the low-resolution feature quantity, and generate an image to which predetermined image processing has been applied Figure 5; Paragraph [0051]: “aggregate feature correspondences between HR region 502 and LR region 504”; Paragraph [0041]: “Finally, each of the enhanced LR regions (e.g., enhanced via the aforementioned feature aggregation processes) may be combined at image reconstruction block 214, resulting in the enhanced output image 216”; and Paragraph [0052]: “a concatenation operation on features in the decoder and their counterparts in the encoder passed through skip connections […] the decoder 514 may map the features into an enhanced output residual image 518 containing the recovered details […] in order to reconstruct the desired HR output”).
Li fails to teach execute attention processing only on the low-resolution feature quantity. However, Srinivas teaches execute attention processing only on the low-resolution feature quantity (Figure 4; Table 1; Page 2, Left Col. 3rd Paragraph: “(1) Use convolutions to efficiently learn abstract and low resolution featuremaps from large images; (2) Use global (all2all) self-attention to process and aggregate the information contained in the featuremaps captured by convolutions […] (2) can deal with large images efficiently by having convolutions do the spatial downsampling and letting attention work on smaller resolutions […] Replace only the final three bottleneck blocks of a ResNet with BoT blocks without any other changes […] in other words, take a ResNet and only replace the final three 3 x 3 convolutions with MHSA layers“; and Page 4, Right Col. 2nd Paragraph: “we believe that the simplest setting that adheres to the above factors would be to incorporate self-attention at the lowest resolution featuremaps in the backbone, i.e., the residual blocks in the c5 stack”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Li to execute attention processing only on the low-resolution feature quantity, as taught by Srinivas, in order to reduce the computational and memory overhead associated with attention processing at larger spatial resolutions while providing global aggregation of information in the feature maps. Srinivas expressly recognizes that the memory and computation required for global self-attention scale quadratically with spatial dimensions and teaches addressing this issue by using convolutions for spatial downsampling and allowing attention to operate on smaller resolution feature maps. This motivation is supported by KSR exemplary rationale (G) Some teaching, suggestion, or motivation in the prior art that would have led one of ordinary skill to modify the prior art reference or to combine prior art reference teachings to arrive at the claimed invention. MPEP 2141 (III).
Regarding claim(s) 2, Li as modified by Srinivas teaches the information processing apparatus according to claim 1, where Li teaches wherein when generating the high-resolution feature quantity, a feature quantity of the largest resolution among a plurality of resolutions of feature quantities is generated (Figure 2; Figure 5; Paragraph [0010]: “first image capture device has a higher quality […] than the second image capture device, while the second FOV […] may have a larger FOV than the first FOV (i.e., the FOV of the image captured by the first image capture device)”; Paragraph [0029] – Paragraph [0030]: “Image 102 represents an input image […] Image 106 represents an input image […] an example of a “low resolution” or “LR” or “lower quality” image”; Paragraph [0039]: “these two images may differ significantly in resolution, sharpness, texture details, noise patterns, subject gestures, and tone appearance, etc.”; and Paragraph [0050]: “Given an LR input image region 504 […] e.g., each with a size of HxWx3, two distinct encoders (506/508) firstly embed them into intermediate feature maps”).
Regarding claim(s) 4, Li as modified by Srinivas teaches the information processing apparatus according to claim 1, where Li teaches wherein the instructions are executed by using neural network (Figure 5; and Paragraph [0050]: “an exemplary neural network architecture 500 for performing reference-based image super-resolution (RefSR) image processing techniques is shown […] Given an LR input image region 504 (e.g., bi-linearly upscaled to match its size to the size of reference image region) and its corresponding HR reference image region 502 (which could, alternatively, be downscaled to match the size of the LR input image, depending on the FOV requirements for the enhanced output image), e.g., each with a size of HxWx3, two distinct encoders (506/508) firstly embed them into intermediate feature maps”).
Regarding claim(s) 16, Li teaches an information processing method for a neural network (Figure 5; and Paragraph [0050]) comprising:
image acquiring to acquire an input image (Figure 2; and Paragraph [0010]: “first image capture device has a higher quality […] than the second image capture device, while the second FOV […] may have a larger FOV than the first FOV (i.e., the FOV of the image captured by the first image capture device));
high-resolution feature quantity encoding to generate at least one high-resolution feature quantity for the input image (Figure 5; and Paragraph [0050]: “[…] HR reference image region 502 […] e.g., each with a size of HxWx3, two distinct encoders (506/508) firstly embed them into intermediate feature maps”);
low-resolution feature quantity encoding to generate a low-resolution feature quantity of a lower resolution from the high-resolution feature quantity (Figure 1; Figure 5; Paragraph [0029] – Paragraph [0030]: “Image 102 represents an input image […] Image 106 represents an input image […] an example of a “low resolution” or “LR” or “lower quality” image”; and Paragraph [0050]: “Given an LR input image region 504 […] e.g., each with a size of HxWx3, two distinct encoders (506/508) firstly embed them into intermediate feature maps”); and
decoding to combine the high-resolution feature quantity and the low-resolution feature quantity, and generate an image to which predetermined image processing has been applied (Figure 5; Paragraph [0051]: “aggregate feature correspondences between HR region 502 and LR region 504”; and Paragraph [0041]: “Finally, each of the enhanced LR regions (e.g., enhanced via the aforementioned feature aggregation processes) may be combined at image reconstruction block 214, resulting in the enhanced output image 216”).
Li fails to teach attention processing to execute attention processing only on the low-resolution feature quantity. However, Srinivas teaches attention processing to execute attention processing only on the low-resolution feature quantity (Figure 4; Table 1; Page 2, Left Col. 3rd Paragraph: “(1) Use convolutions to efficiently learn abstract and low resolution featuremaps from large images; (2) Use global (all2all) self-attention to process and aggregate the information contained in the featuremaps captured by convolutions […] (2) can deal with large images efficiently by having convolutions do the spatial downsampling and letting attention work on smaller resolutions […] Replace only the final three bottleneck blocks of a ResNet with BoT blocks without any other changes […] in other words, take a ResNet and only replace the final three 3 x 3 convolutions with MHSA layers“; and Page 4, Right Col. 2nd Paragraph: “we believe that the simplest setting that adheres to the above factors would be to incorporate self-attention at the lowest resolution featuremaps in the backbone, i.e., the residual blocks in the c5 stack”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Li to execute attention processing only on the low-resolution feature quantity, as taught by Srinivas, in order to reduce the computational and memory overhead associated with attention processing at larger spatial resolutions while providing global aggregation of information in the feature maps. Srinivas expressly recognizes that the memory and computation required for global self-attention scale quadratically with spatial dimensions and teaches addressing this issue by using convolutions for spatial downsampling and allowing attention to operate on smaller resolution feature maps. This motivation is supported by KSR exemplary rationale (G) Some teaching, suggestion, or motivation in the prior art that would have led one of ordinary skill to modify the prior art reference or to combine prior art reference teachings to arrive at the claimed invention. MPEP 2141 (III).
Regarding claim 17, Li teaches a non-transitory computer-readable storage medium configured to store a computer program comprising instructions for executing (Figure 8; and Paragraph [0076]) the following processes:
image acquiring to acquire an input image (Figure 2; and Paragraph [0010]: “first image capture device has a higher quality […] than the second image capture device, while the second FOV […] may have a larger FOV than the first FOV (i.e., the FOV of the image captured by the first image capture device));
high-resolution feature quantity encoding to generate at least one high-resolution feature quantity for the input image (Figure 5; and Paragraph [0050]: “[…] HR reference image region 502 […] e.g., each with a size of HxWx3, two distinct encoders (506/508) firstly embed them into intermediate feature maps”);
low-resolution feature quantity encoding to generate a low-resolution feature quantity of a lower resolution from the high-resolution feature quantity (Figure 1; Figure 5; Paragraph [0029] – Paragraph [0030]: “Image 102 represents an input image […] Image 106 represents an input image […] an example of a “low resolution” or “LR” or “lower quality” image”; and Paragraph [0050]: “Given an LR input image region 504 […] e.g., each with a size of HxWx3, two distinct encoders (506/508) firstly embed them into intermediate feature maps”); and
decoding to combine the high-resolution feature quantity and the low-resolution feature quantity, and generate an image to which predetermined image processing has been applied (Figure 5; Paragraph [0051]: “aggregate feature correspondences between HR region 502 and LR region 504”; and Paragraph [0041]: “Finally, each of the enhanced LR regions (e.g., enhanced via the aforementioned feature aggregation processes) may be combined at image reconstruction block 214, resulting in the enhanced output image 216”).
Li fails to teach attention processing to execute attention processing only on the low-resolution feature quantity. However, Srinivas teaches attention processing to execute attention processing only on the low-resolution feature quantity (Figure 4; Table 1; Page 2, Left Col. 3rd Paragraph: “(1) Use convolutions to efficiently learn abstract and low resolution featuremaps from large images; (2) Use global (all2all) self-attention to process and aggregate the information contained in the featuremaps captured by convolutions […] (2) can deal with large images efficiently by having convolutions do the spatial downsampling and letting attention work on smaller resolutions […] Replace only the final three bottleneck blocks of a ResNet with BoT blocks without any other changes […] in other words, take a ResNet and only replace the final three 3 x 3 convolutions with MHSA layers“; and Page 4, Right Col. 2nd Paragraph: “we believe that the simplest setting that adheres to the above factors would be to incorporate self-attention at the lowest resolution featuremaps in the backbone, i.e., the residual blocks in the c5 stack”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Li to execute attention processing only on the low-resolution feature quantity, as taught by Srinivas, in order to reduce the computational and memory overhead associated with attention processing at larger spatial resolutions while providing global aggregation of information in the feature maps. Srinivas expressly recognizes that the memory and computation required for global self-attention scale quadratically with spatial dimensions and teaches addressing this issue by using convolutions for spatial downsampling and allowing attention to operate on smaller resolution feature maps. This motivation is supported by KSR exemplary rationale (G) Some teaching, suggestion, or motivation in the prior art that would have led one of ordinary skill to modify the prior art reference or to combine prior art reference teachings to arrive at the claimed invention. MPEP 2141 (III).
Regarding claim(s) 18, Li as modified by Srinivas teaches the information processing apparatus according to claim 1, where Li teaches wherein the low-resolution feature quantity is generated by compressing the high-resolution feature quantity that has been generated from the input image (Figure 5; and Paragraph [0050]: “Given an LR input image region 504 […] e.g., each with a size of HxWx3, two distinct encoders (506/508) firstly embed them into intermediate feature maps […] the encoders (506/508) may consist of a set of initial convolution layers with a stride of 1 to extract low-level features from the input region. The following three subsequent convolution layers with a stride of 2 aim to transform the low-level features to more abstract latent space by inflating the number of channels and enlarging the receptive field via downscaling them to feature maps, e.g., of size (H/8) × (W/8) × (Cx8)”), and
where Srinivas teaches wherein the attention processing is executed on the low-resolution feature quantity and the attention processing is not executed on the high-resolution feature quantity (Figure 4; Table 1; Page 2, Left Col. 3rd Paragraph: “(1) Use convolutions to efficiently learn abstract and low resolution featuremaps from large images; (2) Use global (all2all) self-attention to process and aggregate the information contained in the featuremaps captured by convolutions […] (2) can deal with large images efficiently by having convolutions do the spatial downsampling and letting attention work on smaller resolutions […] Replace only the final three bottleneck blocks of a ResNet with BoT blocks without any other changes […] in other words, take a ResNet and only replace the final three 3 x 3 convolutions with MHSA layers“; and Page 4, Right Col. 2nd Paragraph: “we believe that the simplest setting that adheres to the above factors would be to incorporate self-attention at the lowest resolution featuremaps in the backbone, i.e., the residual blocks in the c5 stack”).
Regarding claim(s) 20, Li as modified by Srinivas teaches the information processing apparatus according to claim 1, where Srinivas teaches wherein the at least one processor (Page 5, 4. Experiments, 1st Paragraph) executes the attention processing at an end of a path that generates the low-resolution feature quantity (Table 1; and Page 4, Right Col. 2nd Paragraph: “we believe that the simplest setting that adheres to the above factors would be to incorporate self-attention at the lowest resolution featuremaps in the backbone, i.e., the residual blocks in the c5 stack”),
further compresses the low-resolution feature quantity, on which the attention processing has been executed, to generate a low-resolution feature quantity with a lower resolution (Table 1; and Page 4, 3. Method: “the first MHSA layer of the c5 stack operates on a 64×64 feature map and uses a 2×2 average-pooling operation with a stride of 2 to spatially down sample the feature map, with the subsequent MHSA layers operating on 32×32 feature maps”),
executes the attention processing on each of the low-resolution feature quantities (Table 1; and Page 4, 3. Method: “teaches MHSA processing at the plurality of low-resolution feature-map resolutions, including the first MHSA operating on the 64×64 feature map and the subsequent MHSA layers operating on the further-downsampled 32×32 feature maps”), and
in a decoder, skip-connects the high-resolution feature quantity, on which the attention processing has not been executed, with the low-resolution feature quantities with the plurality of resolutions, on which the attention processing has been executed (Where Li teaches in Figure 5; and Paragraph [0052]: “a concatenation operation on features in the decoder and their counterparts in the encoder passed through skip connections […] the decoder 514 may map the features into an enhanced output residual image 518 containing the recovered details […] in order to reconstruct the desired HR output”; and where Srinivas teaches in Table 1; and Page 4, 3. Method: “restricting self-attention to the low-resolution c5 portion of the network while retaining convolutional processing at the preceding higher-resolution feature-map stages, such that attention processing is not executed on the higher-resolution feature quantity while attention processing is executed on the low-resolution feature quantities at the plurality of low resolutions”).
Claim(s) 3, 5, and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Li et al (US 2023/0098437 A1) in view of Srinivas et al (Bottleneck Transformers for Visual Recognition), further in view of Mo et al (“Frequency Attention Network: Blind Noise Removal for Real Images”).
Regarding claim(s) 3, Li as modified by Srinivas teaches the information processing apparatus according to claim 1, but do not specifically teach wherein in the attention processing, attention processing that weights the input feature quantities in the spatial direction is executed.
However, where Mo teaches wherein in the attention processing, attention processing that weights the input feature quantities in the spatial direction is executed (Figure 3; Abstract: “In this work we propose a method for blind image denoising that combines frequency domain analysis and attention mechanism, named frequency attention network (FAN) […] spatial and channel mechanisms are employed to enhance feature maps at different scales for capturing contextual information”; and Page 7-8, Chapter 3.3 Spatial-Channel Attention Block (SCAB): “the difference between the spatial domain and the value domain to calculate the weight of different locations for the central area, spatial attention mechanism can re-weight the feature map according to the location of the features and help the network learn where to be paid attention”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to further modify Li as modified by Srinivas to utilize Mo's spatial attention mechanism for the attention processing performed on the low-resolution feature quantity, in order to enhance spatially relevant features and improve feature representation by weighting the feature map according to the locations of the features, as taught by Mo. Mo teaches that the spatial attention mechanism “can re-weight the feature map according to the location of the features and help the network learn where to be paid attention.” Thus, utilizing Mo's known spatial attention mechanism for the low-resolution attention processing taught by Srinivas would have predictably enhanced spatially relevant feature information. This motivation is supported by KSR exemplary rationale (G) Some teaching, suggestion, or motivation in the prior art that would have led one of ordinary skill to modify the prior art reference or to combine prior art reference teachings to arrive at the claimed invention. MPEP 2141 (III).
Regarding claim(s) 5, Li as modified by Srinivas teaches the information processing apparatus according to claim 1, but do not specifically teach wherein the predetermined image processing is configured to include noise removal processing. However, Mo teaches wherein the predetermined image processing is configured to include noise removal processing (Abstract: “a method for blind image denoising that combines frequency domain analysis and attention mechanism, named frequency attention network (FAN)”; and Page 6, Chapter 3.2 Wavelet Transform, 1st Paragraph: “Wavelet transformation of one image decomposes the image into different sub-bands based on the frequency information and the processing of the medium and the high frequency sub-bands can result in noise removal”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to further modify Li as modified by Srinivas to perform noise removal processing as taught by Mo, in order to remove noise from an input image and thereby generate an enhanced image with reduced noise. Mo expressly teaches applying its Frequency Attention Network (FAN) to blind image denoising and utilizing frequency-domain information and attention mechanisms to restore a clean image from a noisy image. Therefore, applying Mo's known noise-removal processing to the image processing apparatus of Li as modified by Srinivas would have predictably improved the quality of the processed image by reducing image noise. This motivation is supported by KSR exemplary rationale (G) Some teaching, suggestion, or motivation in the prior art that would have led one of ordinary skill to modify the prior art reference or to combine prior art reference teachings to arrive at the claimed invention. MPEP 2141 (III).
Regarding claim(s) 19, Li as modified by Srinivas teaches the information processing apparatus according to claim 1, where Srinivas teaches wherein the spatial attention processing is not executed on the high-resolution feature quantity (Figure 4; Table 1; Page 2, Left Col. 3rd Paragraph: “(1) Use convolutions to efficiently learn abstract and low resolution featuremaps from large images; (2) Use global (all2all) self-attention to process and aggregate the information contained in the featuremaps captured by convolutions […] (2) can deal with large images efficiently by having convolutions do the spatial downsampling and letting attention work on smaller resolutions […] Replace only the final three bottleneck blocks of a ResNet with BoT blocks without any other changes […] in other words, take a ResNet and only replace the final three 3 x 3 convolutions with MHSA layers“; and Page 4, Right Col. 2nd Paragraph: “we believe that the simplest setting that adheres to the above factors would be to incorporate self-attention at the lowest resolution featuremaps in the backbone, i.e., the residual blocks in the c5 stack”).
Li and Srinivas fail to teach wherein the predetermined image processing includes noise removal processing, wherein the attention processing is spatial attention processing that weights the low- resolution feature quantity in the spatial direction. However, Mo teaches wherein the predetermined image processing includes noise removal processing (Abstract: “a method for blind image denoising that combines frequency domain analysis and attention mechanism, named frequency attention network (FAN)”; and Page 6, Chapter 3.2 Wavelet Transform, 1st Paragraph: “Wavelet transformation of one image decomposes the image into different sub-bands based on the frequency information and the processing of the medium and the high frequency sub-bands can result in noise removal”), wherein the attention processing is spatial attention processing that weights the low- resolution feature quantity in the spatial direction (Figure 3; Abstract: “In this work we propose a method for blind image denoising that combines frequency domain analysis and attention mechanism, named frequency attention network (FAN) […] spatial and channel mechanisms are employed to enhance feature maps at different scales for capturing contextual information”; and Page 7-8, Chapter 3.3 Spatial-Channel Attention Block (SCAB): “the difference between the spatial domain and the value domain to calculate the weight of different locations for the central area, spatial attention mechanism can re-weight the feature map according to the location of the features and help the network learn where to be paid attention”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to further modify Li as modified by Srinivas to utilize Mo's spatial attention mechanism for the attention processing performed on the low-resolution feature quantity and to perform noise removal processing, in order to enhance spatially relevant features and improve image quality by reducing image noise. Mo teaches that its spatial attention mechanism “can re-weight the feature map according to the location of the features and help the network learn where to be paid attention,” and further teaches applying such attention processing in a Frequency Attention Network (FAN) for blind image denoising. Thus, utilizing Mo's spatial attention mechanism for the low-resolution attention processing of Li as modified by Srinivas would have predictably enhanced spatially relevant feature information and facilitated removal of noise from the processed image. This motivation is supported by KSR exemplary rationale (G) Some teaching, suggestion, or motivation in the prior art that would have led one of ordinary skill to modify the prior art reference or to combine prior art reference teachings to arrive at the claimed invention. MPEP 2141 (III).
Relevant Prior Art Directed to State of Art
Nagumo et al (US 2011/0211765 A1) are relevant prior art not applied in the rejection(s) above. Nagumo discloses an image processing apparatus comprising a super resolving processor including: a high frequency estimator which generates difference image information between a low resolution image input as a processing object image of a super resolving process and a mid-processing image of the super resolving process or a processed image, that is, an initial image; and a calculator which performs a process of updating the processed image through a process of calculation between the difference image information output from the high frequency estimator and the processed image, wherein the high frequency estimator performs a learning type data process using learned data in the difference image information generating process.
Narita (US 2016/0260227 A1) are relevant prior art not applied in the rejection(s) above. Narita discloses an image processing apparatus comprising: An image acquiring unit configured to acquire, based on an input image, a plurality of resolution hierarchical images having resolutions that are different from one another; An extracting unit configured to extract a feature point in each of the plurality of resolution hierarchical images; a determining unit configured to determine a correlation among a plurality of feature points in the plurality of resolution hierarchical images using information on positions of the plurality of feature points in each resolution hierarchical image; and a selecting unit configured to select at least one processing object image to be processed with the feature point in the plurality of resolution hierarchical images, based on a determination result of the correlation, wherein at least one processor or circuit is configured to perform a function of at least one of the units.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 JONGBONG NAH whose telephone number is (571)272-1361. The examiner can normally be reached M - F: 7:30am - 4:30pm.
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, Edward Urban can be reached on 571-272-7899. 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.
/JONGBONG NAH/Examiner, Art Unit 2674
/ONEAL R MISTRY/Supervisory Patent Examiner, Art Unit 2674
Prosecution Insights