Image processing for automated object identification

• US 10,706,308 B2
• Filed: 08/07/2018
• Issued: 07/07/2020
• Est. Priority Date: 08/07/2018
• Status: Active Grant

• Current Assignee: Accenture Global Solutions Limited (Accenture PLC)
• Original Assignee: Accenture Global Solutions Limited (Accenture PLC)
• Inventors: Zaman, Md Faisal, Rammos, Perikles, O'Hagan, Cian
• Primary Examiner(s): Tran, Phuoc
• Application Number:
  US16/057,246
• Publication Number:
  US 20200050879A1
• Time in Patent Office: 700 Days
• CPC Class Codes: G05D 1/0246 using a video camera in combination with image processing means
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 2207/30168 Image quality inspection
• CPC Class Codes: G06T 2207/30252 Vehicle exterior; Vicinity of vehicle
• CPC Class Codes: G06T 3/40 Scaling of whole images or parts thereof, e.g. expanding or contracting
• CPC Class Codes: G06T 5/50 using two or more images, e.g. averaging or subtraction
• CPC Class Codes: G06T 7/0002 Inspection of images, e.g. flaw detection
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/82 using neural networks
• CPC Class Codes: G06V 10/993 Evaluation of the quality of the acquired pattern
• CPC Class Codes: G06V 2201/08 Detecting or categorising vehicles

Abstract

A device obtains an image and generates, based on the image and using an image generation model, a set of candidate images for object recognition. The device determines, based on the set of candidate images and using an image evaluation model, that one or more candidate images, of the set of candidate images, are processed during image generation and using the image generation model to improve a resolution of the image. The device determines image embeddings for the one or more candidate images and based on the image, and determines an embeddings error for the one or more candidate images based on the image embeddings and the image. The device determines an identification of a feature of the image based on the image embeddings and the embeddings error, and performs an action relating to the identification of the feature based on determining the identification of the feature.

Claims

1. A device, comprising:

one or more memories; and

one or more processors, communicatively coupled to the one or more memories, to;

obtain an image;

generate, based on the image and using an image generation model, a set of candidate images for object recognition;

determine, based on the set of candidate images and using an image evaluation model, that one or more candidate images, of the set of candidate images, are processed during image generation and using the image generation model to improve a resolution of the image;

determine image embeddings for the one or more candidate images and based on the image;

determine an embeddings error for the one or more candidate images based on the image embeddings and the image;

determine an identification of a feature of the image based on the image embeddings and the embeddings error;

backpropagate the embeddings error into the image generation model to generate an updated image generation model for subsequent image generation;

backpropagate the identification of the feature of the image into the image evaluation model to generate an updated image evaluation model for subsequent image evaluation; and

perform an action relating to the identification of the feature based on determining the identification of the feature.

2. The device of claim 1, wherein the one or more processors are further to:

use the updated image generation model to generate another set of candidate images based on another image; and

determine another identification of another feature of the other image based on the other set of candidate images.

3. The device of claim 1, wherein the one or more processors are further to:

use the updated image evaluation model to evaluate another one or more candidate images relating to another image; and

determine another identification of another feature of the other image based on the other one or more candidate images.

4. The device of claim 1, wherein the one or more processors are further to:

determine that the embeddings error does not satisfy a threshold;

iterate, based on determining that the embeddings error does not satisfy the threshold, image generation and image evaluation using a result of determining the image embeddings and the embeddings error to determine another image embeddings and another embeddings error; and

wherein the one or more processors, when determining the identification of the feature of the image, are to;

determine the identification of the feature of the image based on the other image embeddings and the other embeddings error.

5. The device of claim 1, wherein the action is associated with causing an autonomous vehicle to be controlled.

6. The device of claim 1, wherein the one or more processors, when performing the action relating to the identification of the feature, are to:

communicate with another device to provide a command relating to altering functioning of the other device based on the identification of the feature.

7. The device of claim 1, wherein the feature of the image is at least one of:

a biometric feature,

an object,

an image segment, or

a message.

8. The device of claim 1, wherein the image evaluation model is a deep reinforcement learning model.

9. The device of claim 1, wherein the image generation model is a generative adversarial networks (GANs) model.

10. The device of claim 1, wherein the image is at least one frame of a video recording.

11. The device of claim 1, wherein the image comprises a low resolution image.

12. The device of claim 1, wherein the one or more processors, when determining, using the image evaluation model, that the one or more candidate images are processed during image generation and using the image generation model, are to:

receive a set of historical high resolution images; and

compare the one or more candidate images to the set of historical high resolution images.

13. A method, comprising:

obtaining, by a device, an item of media;

generating, by the device and based on the item of media and using a media generation model, a set of candidate media for object recognition;

determining, by the device and based on the set of candidate media and using a media evaluation model, that one or more candidate media, of the set of candidate media, are processed during media generation and using the media generation model to improve a quality of a feature of the item of media;

determining, by the device, media embeddings for the one or more candidate media and based on the item of media;

determining, by the device, an embeddings error for the one or more candidate media based on the media embeddings and the item of media;

determining, by the device, an identification of the feature of the item of media based on the media embeddings and the embeddings error; and

communicating, by the device, with another device to provide information identifying the feature based on determining the identification of the feature.

14. The method of claim 13, wherein the item of media is at least one of:

an image,

a video recording,

an audio recording,

a virtual reality representation, or

a data representation.

15. The method of claim 13, wherein the quality of the feature is at least one of:

a resolution,

a sharpness, or

a contrast.

16. The method of claim 13, wherein determining the identification of the feature comprises:

determining the identification of the feature using a simultaneous localization and mapping (SLAM) object detection procedure.

17. The method of claim 13, wherein determining the identification of the feature comprises:

identifying an object in the item of media.

18. The method of claim 13, further comprising:

segmenting the item of media based on the identification of the feature; and

wherein communicating with the other device to provide information identifying the feature comprises;

providing information relating to segmenting the item of media.

19. The method of claim 13, wherein obtaining the item of media comprises:

requesting a low-resolution version of the item of media,

wherein the low-resolution version is associated with a reduced file-size relative to other versions of the item of media; and

wherein generating the set of candidate media comprises;

generating at least one high-resolution version of the item of media based on the low-resolution version,

wherein the at least one high-resolution version is associated with a higher file-size than the low-resolution version.

20. A non-transitory computer-readable medium storing instructions, the instructions comprising:

one or more instructions that, when executed by one or more processors, cause the one or more processors to;

generate, based on an image and using an image generation model, a set of candidate images for object recognition,

wherein each candidate image, in the set of candidate images, is associated with an increased image resolution relative to the image;

determine image embeddings for the set of candidate images and based on the image;

determine that an embeddings error threshold for the image embeddings is satisfied;

determine an identification of a feature of the image based on the image embeddings and based on determining that the embeddings error threshold is satisfied; and

provide information identifying the feature of the image based on determining the identification of the feature.

21. The non-transitory computer-readable medium of claim 20, wherein the one or more instructions, when executed by the one or more processors, further cause the one or more processors to:

use a reinforcement learning based image evaluation module to evaluate the set of candidate images;

determine that a resolution criterion is satisfied based on evaluating the set of candidate images; and

wherein the one or more instructions, that cause the one or more processors to determine the image embeddings, cause the one or more processors to;

determine the image embeddings based on determining that the resolution criterion is satisfied.

22. The non-transitory computer-readable medium of claim 20, wherein the image is a video and wherein the set of candidate images is a set of frames of the video.