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
Acknowledgment is made of applicant’s claim for foreign priority under 35 U.S.C. 119 (a)-(d). The certified copy has been filed in parent Application No. KR10-2023-0118490, filed on 09/06/2023; as well as Application No. KR10-2022-0114099, filed on 09/08/2022.
Response to Amendment
The amendment filed on 06/23/2026 has been entered.
Claims 1, 7, 8, 14-18, 20 were amended.
Claims 6, 13, and 19 were cancelled.
Claims 1-5, 7-12, 14-18, 20 remain pending in the application.
The specification objection to the title has been removed due to the amendments.
Response to Arguments
Applicant’s arguments (Remarks filed 06/23/2026) have been considered but are not fully persuasive.
Applicant has made the following arguments on pg 10 (Remarks filed 06/23/2026), first full paragraph, reproduced below:
PNG
media_image1.png
500
978
media_image1.png
Greyscale
.
Upon further review of the reference and in light of applicant's argument, the examiner respectfully disagrees as follows: first of all, Cui’s experiment on how scaling affects reconstruction quality is later used in the gain unit (Section 3.1 and 3.2) as a way to control quality (or information loss). Applicant states that this is different from deriving a set of selected elements of a quantized latent representation and entropy encoding the selected elements. Figure 6 demonstrates the process; starting in the top-left area, the “Attention Module”, as one with ordinary skill in the art would know, derives selected elements in the latent representation using attention (involving importance, or saliency, mapping or masking). Part of this data is then passed to the Gain Unit (which helps in controlling quality), then passed into the UnivQuan (which converts the latent representation into a quantized latent representation). The instant claims state “deriving a set of selected elements of the quantized latent representation”; the attention module points out important elements, or selected elements with the importance map; then at the UnivQuan stage, the selected elements are selected, or derived, at this point. After the UnivQuan stage, these (which includes the selected elements) are passed into a process that involves the Asymmetry Gaussian Entropy Model, which handles entropy encoding (or entropy coding); do note there is another entropy model in Figure 6, along with encoding/decoding sections. In summary, the gain unit helps with controlling quality (or information loss), which includes the selected elements (because of the importance map), and then entropy encoding is applied.
Applicant makes further arguments on pg 10, last two paragraphs, reproduced below:
PNG
media_image2.png
438
962
media_image2.png
Greyscale
.
Upon further review of the reference and in light of applicant's argument, the examiner respectfully disagrees as follows: first of all, as stated earlier, and with Figure 6, the gain unit is used, which is then passed into quantization, and then passed into entropy encoding. The gain unit is part of the process of entropy encoding.
Regarding claims 3, 10, and 17, Applicant makes the following argument on pg 12, paragraph 1, reproduced below:
PNG
media_image3.png
390
966
media_image3.png
Greyscale
Upon further review of the reference and in light of applicant's argument, the examiner respectfully disagrees as follows: first of all, Applicant appears to summarize the Li method and state that Li does not appear to have the limitations from claim 1. Evidence doesn’t appear to be shown, but hinted at, similar as to the why; in the interest of compact prosecution, Examiner will address some of the statements. Applicant’s summary of Li discloses selected elements of the quantized latent representation (“which portions of binary codes are retained or trimmed”); “binary codes” are the quantized latent representations, “which portions” shows selected elements. Entropy coding (or entropy encoding), is taught in Section 3.1.3 (¶1) of Li: “the code length after quantization is spatially invariant, and entropy coding is then used”. Further, “quantized latent representation” doesn’t appear to be specifically defined in the specifications, nor in the claims. One with ordinary skill in the art would guess that these are the binary codes so that entropy coding may be done to encode and predict the entropy pattern in the binary code, otherwise, entropy coding would not work at all.
The remaining dependent claims rely upon the independent claims, and are rejected for similar reasons.
Accordingly, the claims, as they are currently written, do not place the instant application into an allowable state.
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.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1, 2, 7-9, 14-16, 20 is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Cui (“Asymmetric Gained Deep Image Compression With Continuous Rate Adaptation”, Aug 2022).
Regarding claims 1, 8, and 15, Cui teaches A method for image encoding (Cui, pg 1, column 2, first full paragraph, reproduced below:
PNG
media_image4.png
284
760
media_image4.png
Greyscale
, “Novel image compression framework, AG-VAE” is being interpreted as involving image encoding [for claim 1] and decoding [for claims 8 and 15]), comprising:
generating a latent representation using an input image (Cui, pg 3, column 1, Section 3.1, ¶1, reproduced below:
PNG
media_image5.png
764
724
media_image5.png
Greyscale
. “We taken one image as input of the encoder to obtain its latent representation”);
generating a quantized latent representation (Cui, pg 3, column 1, Section 3.1, ¶2-3, reproduced below:
PNG
media_image6.png
674
742
media_image6.png
Greyscale
. “Quantization loss of the latent representation” is being interpreted as involving “generating a quantized latent representation”) by performing adaptive quantization on the latent representation (Cui, pg 3, column 1, see Section 3.1, ¶2-3 image above: “scale the latent representation flexibly” is being interpreted as involving “adaptive quantization on the latent representation”);
deriving a set of selected elements of the quantized latent representation (Cui, see Section 3.1 images above for full context, but ¶1 cites: “We can conclude that the channels’ importance varies and can be scaled to control the reconstruction quality”. “Importance varies” shows a set of selected elements, the importance values. When combined with the “scale the latent representation flexibly” from ¶2 and the “quantization loss” from ¶3, shows “quantized latent representation”); and
generating encoded information of the selected elements by performing entropy encoding on the set of the selected elements (Cui, pg 1 column 2 last paragraph to pg 2 column lines1-4, reproduced below:
PNG
media_image7.png
529
1142
media_image7.png
Greyscale
. “Asymmetric Gaussian entropy model” is being interpreted as involving entropy encoding. As the Cui framework does encoding and decoding, as see in Figure 1 text. Further. Section 3.5 shows the entropy model generates encoded information [“estimate the distribution of the latent representation”]), wherein
wherein the encoded information of the selected elements (Cui, pg 5, Fig 6, “Attention Module”, which is mentioned in reference 36, uses an importance map that obtains selected elements) is generated using an entropy-model (Cui, Figure 6, “Asymmetry Gaussian Entropy Model” is being interpreted as an example entropy-model that encodes information) parameter for a specific target quality level (Cui, pg 3, Section 3.2, ¶1: “By scaling the y to different intervals channel wisely, the gain unit can adjust the channel redundancy, thus control information loss of the quantization process effectively”. “Information loss” is being interpreted to involve quality, as the scaling affects reconstruction quality (Section 3.1). “Controlling information loss” is being interpreted to involve a “parameter for a specific target quality level”).
Regarding claim 2, Cui teaches The method of claim 1, wherein the quantized latent representation is generated for a specific target quality level (Cui, pg 3, section 3.1, ¶1, reproduced below:
PNG
media_image8.png
804
758
media_image8.png
Greyscale
“can be scaled to control the reconstruction quality” is being interpreted to involve “a specific target quality level”).
Regarding claim 7, Cui teaches The method of claim 1, wherein the entropy-model parameter includes a scale parameter for the specific target quality level (Cui, section 3.1, ¶1, reproduced below:
PNG
media_image9.png
802
758
media_image9.png
Greyscale
. “can be scaled to control the reconstruction quality” is being interpreted as “a specific target quality level”. The “channels’ importance varies” is being interpreted as “a parameter” that can be scaled) or a mean parameter for the specific target quality level.
Claims 9 and 16 is/are rejected using the same rationale as applied to claim 2 discussed above.
Claims 14 and 20 is/are rejected using the same rationale as applied to claim 7 discussed above.
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.
Claim(s) 3, 10, 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Cui, in view of Li (“Learning Convolutional Networks for Content-weighted Image Compression”, 2017).
Regarding claim 3, Cui teaches The method of claim 1, wherein the set of the selected elements (Cui, see Section 3.1 images above for full context, but ¶1 cites: “We can conclude that the channels’ importance varies and can be scaled to control the reconstruction quality”. “Importance varies” shows a set of selected elements, the importance values. When combined with the “scale the latent representation flexibly” from ¶2 and the “quantization loss” from ¶3, shows “quantized latent representation”)
However, Cui does not appear to explicitly teach 3D binary mask; though does teach importance values as the Li reference does.
Pertaining to the same field of endeavor, Li teaches
is determined using a 3D binary mask (Li, pg 4, column 2, full paragraphs 1-3, reproduced below:
PNG
media_image10.png
726
556
media_image10.png
Greyscale
. The “importance mask” with three-dimensions n x h x w, is being interpreted as involving a “3D binary mask”. Equation 5 shows the binary nature of the importance mask).
Cui and Li are considered to be analogous art because they are directed to learned image compression. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method and system for image compression with importance values (as taught by Cui) to include a 3D binary mask (as taught by Li) because the combination provides an improvement to image compression (Li, abstract). Further, Cui teaches 3D latent representation (Cui, section 3.1, line 5).
Claims 10 and 17 is/are rejected using the same rationale as applied to claim 3 discussed above.
Claim(s) 4, 5, 11, 12, 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Cui, as modified by Li, in view of Balle (“Variational image compression with a scale hyperprior”, 2018).
Regarding claim 4, Cui teaches The method of claim 3, wherein the 3D binary mask is generated (Li, pg 4, column 2, full paragraphs 1-3, reproduced below:
PNG
media_image10.png
726
556
media_image10.png
Greyscale
. The “importance mask” with three-dimensions n x h x w, is being interpreted as involving a “3D binary mask”. Equation 5 shows the binary nature of the importance mask).
However, Cui and Li do not appear to specifically teach hyper-decoder.
Pertaining to the same field of endeavor, Balle teaches
using output of a specific layer of a hyper-decoder (Balle, pg 6, Figure 4, reproduced below:
PNG
media_image11.png
630
918
media_image11.png
Greyscale
”arithmetic decoder” is being interpreted as hyper-decoder since the right side of Figure 4 is the hyperprior model. The hyperprior model is being interpreted as outputting, note the z leading to the Q out of the specific layer from the hyperprior model at the end).
Cui, Li, and Balle are considered to be analogous art because they are directed to learned image compression. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method and system for image compression with importance values, modified to have a 3D binary mask (as taught by Cui and Li) to include an output of a specific layer of a hyper-decoder (as taught by Balle) because the combination provides an improvement to image compression (Balle, abstract).
Regarding claim 5, Balle teaches The method of claim 4, wherein a hyperprior is input to the hyper-decoder (Balle, pg 6, Figure 4, reproduced below:
PNG
media_image11.png
630
918
media_image11.png
Greyscale
”arithmetic decoder” is being interpreted as hyper-decoder since the right side of Figure 4 is the hyperprior model. The hyperprior is input into the hyper-decoder on the right side).
Cui, Li, and Balle are considered to be analogous art because they are directed to learned image compression. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method and system for image compression with importance values, modified to have a 3D binary mask (as taught by Cui and Li) to include the hyperprior as input to the hyper-decoder (as taught by Balle) because the combination provides an improvement to image compression (Balle, abstract).
Claims 11 and 18 is/are rejected using the same rationale as applied to claim 4 discussed above.
Claim 12 is/are rejected using the same rationale as applied to claim 5 discussed above.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Zhang et al (“Residual Non-Local Attention Networks for Image Restoration”, 2019) discloses entropy encoding of selected elements (interpreted from attention networks).
Qian et al (‘Entroformer: A Transformer-Based Entropy Model for Learned Image Compression”, 2022) discloses entropy encoding of selected elements (interpreted from attention networks).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOHNNY B DUONG whose telephone number is (571)272-1358. The examiner can normally be reached Monday - Thursday 10a-9p (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, Matthew Bella can be reached at (571)272-7778. 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.
/J.B.D./Examiner, Art Unit 2667
/MATTHEW C BELLA/Supervisory Patent Examiner, Art Unit 2667
Prosecution Insights