Neural Image Decoding with Hyperprior-Guided Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing encoding and decoding methods based on neural networks suffer from poor performance, high complexity, and inefficient utilization of hyperprior side information in end-to-end image coding frameworks, limiting the quality of reconstructed images.
Innovation Solution
A decoding method that enhances features using probability distribution parameters and enhancement parameters encoded in header information bitstreams, and an encoding method that encodes these parameters to improve the quality of reconstructed images without altering primary information, thereby optimizing encoding and decoding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If neural networks are used for encoding and decoding pictures, then performance potential is improved, but encoding performance, decoding performance deteriorate and complexity increases
Solution Approach 1:
The patent segments the neural network processing into two independent parts: a neural network encoder that processes picture features and a neural network decoder that reconstructs images. This segmentation allows each component to be optimized independently, improving overall reliability while maintaining the performance benefits of neural networks.
Solution Approach 2:
The patent introduces hyperprior side information as an intermediary element that bridges the encoder and decoder. This hyperprior information is encoded separately and used to enhance the decoding process, acting as a mediator that improves reconstruction quality without requiring the entire neural network to be retrained for each specific task.
2Adaptability or versatility
If neural networks are used for encoding and decoding pictures, then performance potential is improved, but complexity increases
Solution Approach 1:
By segmenting the system into independent encoder and decoder components with separate neural networks, the patent reduces the complexity of training and deployment. Each component can be trained independently on specialized data, reducing the overall computational burden compared to a single integrated neural network system.
Solution Approach 2:
The patent changes the parameters of the neural networks by introducing hyperprior side information that modifies the probability distribution parameters of the encoded features. This allows the system to adapt to different picture types and qualities without requiring complex retraining, reducing operational complexity while maintaining high performance.
3Productivity
If hyperprior side information is utilized in end-to-end image coding frameworks, then encoding efficiency is improved, but quality of reconstructed images deteriorates
Solution Approach 1:
The patent uses hyperprior side information as an intermediary that enhances the decoding process. This hyperprior information is encoded efficiently but is then used to refine the reconstructed images through the neural network decoder, thereby improving image quality while maintaining encoding efficiency.
Solution Approach 2:
The patent implements a feedback mechanism where the hyperprior side information is used to adjust and refine the reconstruction process. The decoder uses the hyperprior information to guide the reconstruction, providing feedback that improves image quality while maintaining the efficiency benefits of the encoding process.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Provided in the present application are a decoding method and apparatus, a coding method and apparatus, and devices. The decoding method comprises: decoding a code stream corresponding to the current image block, so as to obtain a coefficient hyper-parameter feature corresponding to the current image block; determining a probability distribution parameter on the basis of the coefficient hyper-parameter feature; on the basis of the probability distribution parameter, decoding a code stream corresponding to the current image block, so as to obtain an initial reconstruction feature corresponding to the current image block; and on the basis of the initial reconstruction feature, determining a target reconstructed image block corresponding to the current image block. By means of the technical solution of the present application, the coding performance and the decoding performance can be improved.