# US 11,443,057 B2 · Encoding and concealing information using deep learning > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,443,057 B2 - Document ID: US11443057B2 - Status: Granted - Kind: B2 - Application number: 16/662,779 - Filed: 2019-10-24 - Published: 2021-04-29 - Granted: 2022-09-13 - Earliest claimed: 2019-10-24 - Assignees: AT&T Intellectual Property I, L.P. - Applicants: AT&T Intellectual Property I, L.P. - Inventors: Wei Wang; Mikhail Istomin - Web page: https://patentgrep.com/patent/US11443057B2 ## CPC - G06Q 10/40: Business processes related to social networking or social networking services - G06F 17/16: Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)} - G06F 21/6209: {to a single file or object, e.g. in a secure envelope, encrypted and accessed using a key, or with access control rules appended to the object itself} - G06F 21/6227: {where protection concerns the structure of data, e.g. records, types, queries} - G06N 3/045: Combinations of networks - G06N 3/0455: Auto-encoder networks; Encoder-decoder networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/08: Learning methods - G06N 3/088: Non-supervised learning, e.g. competitive learning - G06N 3/09: Supervised learning - G06T 5/50: using two or more images, e.g. averaging or subtraction - G06T 1/0021: {Image watermarking} - G06T 2207/20081: Training; Learning - G06T 2207/20084: Artificial neural networks [ANN] - G06T 2207/20221: Image fusion; Image merging - G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL ## Abstract Aspects of the subject disclosure may include, for example, a method for training a deep learning model that includes encoding a content item; generating a blended image by combining a background image and the encoded content; decoding the blended image to generate decoded content corresponding to the content item; and defining or specifying a loss function related to the deep learning model. The method also includes determining values of training parameters for the deep learning model to minimize the loss function, thereby obtaining a trained deep learning model. The method also includes an information concealing procedure using the trained deep learning model to conceal user content by encoding the user content and blending the encoded user content with a user-selected image; the information concealing procedure is substantially independent of the user-selected image. Other embodiments are disclosed. ## Claims ### Claim 1 A method comprising: performing, by a processing system including a processor, a training procedure for a deep learning model, the training procedure comprising: encoding a first content item, thereby generating first encoded content, generating a first blended image by combining a first background image and the first encoded content, calculating a first calculated loss determined by a first matrix representing the first background image and a second matrix representing the first blended image, decoding the first blended image, thereby generating first decoded content corresponding to the first content item, calculating a second calculated loss determined by a third matrix representing the first content item and a fourth matrix representing the first decoded content, and a loss function determined in accordance with the first calculated loss and the second calculated loss, and determining values of parameters for the deep learning model to minimize the loss function, thereby obtaining a trained deep learning model; and performing, by the processing system, an information concealing procedure using the trained deep learning model, the information concealing procedure comprising: encoding a second content item, thereby generating second encoded content, and generating a second blended image by combining a second background image and the second encoded content, wherein the information concealing procedure is substantially independent of the second background image. ### Claim 2 (depends on 1) The method of claim 1, wherein the decoding the first blended image comprises extracting the first encoded content from the first blended image and decoding the first encoded content, wherein the decoding is performed using a portion of the deep learning model. ### Claim 3 (depends on 2) The method of claim 2, wherein the portion of the deep learning model comprises a convolutional decoder that is publicly accessible. ### Claim 4 (depends on 1) The method of claim 1, wherein the loss function is calculated as a convex combination of the first calculated loss and the second calculated loss. ### Claim 5 (depends on 1) The method of claim 1, wherein the generating the first blended image comprises performing an element-wise combination of the first matrix and an encoded content matrix representing the first encoded content. ### Claim 6 (depends on 1) The method of claim 1, wherein the first calculated loss corresponds to a distortion of the first background image due to generating the first blended image, and wherein the second calculated loss corresponds to a reconstruction error determined by comparing the first content item with the first decoded content. ### Claim 7 (depends on 1) The method of claim 1, wherein in the training procedure the first background image is randomly selected. ### Claim 8 (depends on 1) The method of claim 1, wherein the second content item and the second background image are user-selected, wherein the generating the second blended image results in a secure image concealing the second content item, and wherein the information concealing procedure further comprises facilitating sharing the secure image with a recipient. ### Claim 9 (depends on 8) The method of claim 8, wherein the sharing is performed via a social network. ### Claim 10 (depends on 1) The method of claim 1, wherein the first content item and the second content item comprise a same content. ### Claim 11 A device, comprising: a processing system including a processor; and a memory that stores executable instructions that, when executed by the processing system, facilitate performance of operations, the operations comprising: performing a training procedure for a deep learning model, the training procedure comprising: encoding a first content item, thereby generating first encoded content, generating a first blended image by combining a first background image and the first encoded content, wherein the first background image is randomly selected, calculating a first calculated loss determined by a first matrix representing the first background image and a second matrix representing the first blended image, decoding the first blended image, thereby generating decoded content corresponding to the first content item, calculating a second calculated loss determined by a third matrix representing the first content item and a fourth matrix representing the decoded content, and a loss function determined in accordance with the first calculated loss and the second calculated loss, and determining values of parameters for the deep learning model to minimize the loss function, thereby obtaining a trained deep learning model; and performing an information concealing procedure using the trained deep learning model, the information concealing procedure comprising: encoding a second content item, thereby generating second encoded content, and generating a second blended image by combining a second background image and the second encoded content, wherein the information concealing procedure is substantially independent of the second background image. ### Claim 12 (depends on 11) The device of claim 11, wherein the decoding the first blended image comprises extracting the first encoded content from the first blended image and decoding the first encoded content, wherein the decoding is performed using a portion of the deep learning model. ### Claim 13 (depends on 11) The device of claim 11, wherein the loss function is calculated as a convex combination of the first calculated loss and the second calculated loss. ### Claim 14 (depends on 11) The device of claim 11, wherein the generating the first blended image comprises performing an element-wise combination of the first matrix and an encoded content matrix representing the first encoded content. ### Claim 15 (depends on 11) The device of claim 11, wherein the first calculated loss corresponds to a distortion of the first background image due to generating the first blended image, and wherein the second calculated loss corresponds to a reconstruction error determined by comparing the first content item with the decoded content. ### Claim 16 A non-transitory machine-readable medium comprising executable instructions that, when executed by a processing system including a processor, facilitate performance of operations, the operations comprising: performing a training procedure for a deep learning model, the training procedure comprising: encoding a first content item, thereby generating first encoded content, generating a first blended image by combining a first background image and the first encoded content, calculating a first calculated loss determined by a first matrix representing the first background image and a second matrix representing the first blended image, decoding the first blended image, thereby generating decoded content corresponding to the first content item, calculating a second calculated loss determined by a third matrix representing the first content item and a fourth matrix representing the decoded content, and a loss function determined in accordance with the first calculated loss and the second calculated loss, and determining values of parameters for the deep learning model to minimize the loss function, thereby obtaining a trained deep learning model; and performing an information concealing procedure using the trained deep learning model, the information concealing procedure comprising: encoding a second content item, thereby generating second encoded content, and generating a second blended image by combining a user-selected second background image and the second encoded content, wherein the information concealing procedure is substantially independent of the second background image. ### Claim 17 (depends on 16) The non-transitory machine-readable medium of claim 16, wherein the decoding the first blended image comprises extracting the first encoded content from the first blended image and decoding the first encoded content, wherein the decoding is performed using a portion of the deep learning model. ### Claim 18 (depends on 16) The non-transitory machine-readable medium of claim 16, wherein the loss function is calculated as convex combination of the first calculated loss and the second calculated loss. ### Claim 19 (depends on 16) The non-transitory machine-readable medium of claim 16, wherein the generating the first blended image comprises performing an element-wise combination of the first matrix and an encoded content matrix representing the first encoded content. ### Claim 20 (depends on 16) The non-transitory machine-readable medium of claim 16, wherein the first calculated loss corresponds to a distortion of the first background image due to generating the first blended image, and wherein the second calculated loss corresponds to a reconstruction error determined by comparing the first content item with the decoded content.