End-to-end joint optimization method for image compression transmission application

By building a feature extraction and reconstruction network in a low-bandwidth image transmission system and combining deep learning with traditional coding techniques for end-to-end joint optimization, the contradiction between software and hardware and the problem of gradient nondifferentiability are solved, improving image compression efficiency and quality, and making it particularly suitable for image transmission in low-bandwidth environments.

CN119767041BActive Publication Date: 2025-10-24XIDIAN UNIV
View PDF -1 Cites 0 Cited by

Patent Information

Application Number
CN202411852671.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-12-16
Publication Date
2025-10-24
Estimated Expiration
2044-12-16

AI Technical Summary

Technical Problem

In existing technologies, the contradiction between hardware and software leads to problems such as image compression algorithms being unable to meet computational requirements on hardware and difficulties in iterative optimization, and neural network gradients being unable to propagate effectively in the backpropagation direction.

Method used

By building a feature extraction and reconstruction network in a low-bandwidth image transmission system, designing a simulated encoding and decoding neural network, and adopting a fixed parameter training strategy, combined with deep learning and traditional image coding techniques, end-to-end joint optimization was performed to solve the non-differentiability problem and improve image compression efficiency.

Benefits of technology

It improves the compression coding efficiency during image transmission, effectively removes spatial redundancy, improves image quality in low-bandwidth environments, resolves hardware and software conflicts, and preserves image edge and detail information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119767041B_ABST
    Figure CN119767041B_ABST
Patent Text Reader

Abstract

The application provides an end-to-end joint optimization method for image compression transmission application, combines deep learning with traditional image coding technology, utilizes different training strategies to gradually optimize the neural network, solves the problem that the gradient of the neural network is difficult to back propagate due to non-differentiability, improves the current situation that similar compatible image codec only restores compressed images from the perspective of post-processing, no longer regards the image compression process as an image degradation problem through a fixed image prior model, improves the compression coding efficiency in the image transmission process, removes the spatial redundancy in the image to a greater extent, effectively solves the problem of image quality decline in a low-bandwidth environment, and can be applied to image compression transmission.
Need to check novelty before this filing date? Find Prior Art