Neural network-based image compression method

Patent metadata

Field	Value
Publication number	US-12475602-B2
Application number	US-202318338613-A
Country	US
Kind code	B2
Filing date	Jun 21, 2023
Priority date	Jun 22, 2022
Publication date	Nov 18, 2025
Grant date	Nov 18, 2025

Abstract

A method of learning a neural network-based image compression model according to the disclosed embodiment, may include receiving a learning target image as an input; encoding the input image through the neural network-based image compression model configured to include a weight parameter, and decoding the encoded image through the neural network-based image compression model; calculating an entropy estimation value for a network model weight of the neural network-based image compression model; calculating a reconstruction performance value by comparing qualities of the learning target image and the decoded image; and learning the neural network-based image compression model by updating the weight parameter of the neural network-based image compression model based on the entropy estimation value for the network model weight and the reconstruction performance value. Accordingly, it is possible to minimize the size of the weight of the neural network-based image compression model.

First claim

What is claimed is:

1 . A method of learning a neural network-based image compression model comprising:
receiving a learning target image as an input; encoding the input image through the neural network-based image compression model configured to include a weight parameter, and decoding the encoded image through the neural network-based image compression model; calculating an entropy estimation value for a network model weight of the neural network-based image compression model; calculating a reconstruction performance value by comparing qualities of the learning target image and the decoded image; and learning the neural network-based image compression model by updating the weight parameter of the neural network-based image compression model based on the entropy estimation value for the network model weight and the reconstruction performance value.

2 . The method of claim 1 , wherein the entropy estimation value for the network model weight is derived to be minimized.

3 . The method of claim 2 , wherein a substitution function to which a uniform distribution is added is used to minimize the entropy estimation value for the network model weight.

4 . The method of claim 3 , wherein the substitution function to which the uniform distribution is added is to add a uniform distribution in a range of −0.5 to 0.5 to the network model weight by an approximate estimation method.

5 . The method of claim 1 , wherein the update of the weight parameter of the neural network-based image compression model is performed by adjusting the entropy estimation value for the network model weight through the parameter.

6 . An encoding method using a method of learning a neural network-based image compression model comprising:
receiving, by an encoder, a compressing target image as an input, and; generating, by the encoder, streaming data by encoding the input compressing target image through the neural network-based image compression model, wherein the neural network-based image compression model is a model that receives a learning target image as an input, encodes the learning target image through the neural network-based image compression model configured to include a weight parameter, and decodes the encoded image through the neural network-based image compression model, calculates an entropy estimation value for a network model weight of the neural network-based image compression model, calculates a reconstruction performance value by comparing qualities of the learning target image and the decoded image, and learns by updating the weight parameter of the neural network-based image compression model based on the entropy estimation value for the network model weight and the reconstruction performance value.

7 . The method of claim 6 , wherein in the neural network-based image compression model, the entropy estimation value for the network model weight is derived to be minimized, a substitution function to which a uniform distribution is added is used to minimize the entropy estimation value for the network model weight, and the substitution function to which the uniform distribution is added is to add a uniform distribution in a range of −0.5 to 0.5 to the network model weight by an approximate estimation method.

8 . The method of claim 6 , wherein the update of the weight parameter of the neural network-based image compression model is performed by adjusting the entropy estimation value for the network model weight through the parameter.

9 . A decoding method using a method of learning a neural network-based image compression model comprising:
receiving, by a decoder, an encoded streaming data as an input through an encoder; and decoding, by the decoder, the streaming data through the neural network-based image compression model, wherein the neural network-based image compression model is a model that receives a learning target image as an input, encodes the learning target image through the neural network-based image compression model configured to include a weight parameter, and decodes the encoded image through the neural network-based image compression model, calculates an entropy estimation value for a network model weight of the neural network-based image compression model, calculates a reconstruction performance value by comparing qualities of the learning target image and the decoded image, and learns by updating the weight parameter of the neural network-based image compression model based on the entropy estimation value for the network model weight and the reconstruction performance value.

10 . The method of claim 9 , wherein in the neural network-based image compression model, the entropy estimation value for the network model weight is derived to be minimized, a substitution function to which a uniform distribution is added is used to minimize the entropy estimation value for the network model weight, and the substitution function to which the uniform distribution is added is to add a uniform distribution in a range of −0.5 to 0.5 to the network model weight by an approximate estimation method.

11 . The method of claim 9 , wherein the update of the weight parameter of the neural network-based image compression model is performed by adjusting the entropy estimation value for the network model weight through the parameter.

Assignees

• Research & Business Found Sungkyunkwan Univ

Inventors

• Ryu Eun-Seok
• Lee Soonbin
• Jeong Jong-Beom

Classifications

• G06T2207/30168

  Image quality inspection · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T7/0002

  Inspection of images, e.g. flaw detection · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title