Neural-Network Post-Filter Signaling in Image Decoding Bitstreams
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing demand for high-resolution and high-quality images leads to a significant increase in transmission and storage costs due to the increased amount of transmitted information or bits, necessitating high-efficient image compression technology.
Innovation Solution
An image encoding/decoding method that includes determining a neural-network post-filter (NNPF) based on supplemental enhancement information (SEI) messages and encoding/decoding whether a target neural-network post-processing filter is activated, using a neural-network post-filter characteristics (NNPFC) SEI message to improve encoding/decoding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If high-resolution and high-quality images are transmitted, then image quality is improved, but transmission cost and storage cost increase
Solution Approach 1:
The patent extracts and transmits only the essential image information through efficient compression algorithms, separating critical visual data from redundant information. This allows high-quality image reconstruction with reduced bitstream volume, directly addressing the contradiction between image quality and transmission amount.
Solution Approach 2:
The patent employs advanced compression parameters and neural-network post-filters that dynamically adjust encoding parameters to maintain perceptual image quality while minimizing the amount of transmitted data. By changing how information is represented rather than simply reducing quantity, the system achieves both goals.
2Adaptability or versatility
If multiple neural-network post-filters are available, then image processing flexibility is improved, but difficulty in detecting and measuring which filter is applied increases
Solution Approach 1:
The patent implements a feedback mechanism where the encoder signals which neural-network post-filter is applied through SEI messages in the bitstream. The decoder receives this feedback information and applies the corresponding filter, ensuring both encoder and decoder use the same filter. This resolves the identification difficulty while maintaining filter selection flexibility.
Solution Approach 2:
The patent introduces SEI (Supplemental Enhancement Information) messages as an intermediary layer between the encoder and decoder. These messages carry metadata about which neural-network post-filter is applied, acting as a mediator that communicates filter selection information without affecting the core image data transmission.
3Reliability
If neural-network post-filter activation information is encoded, then encoding completeness is improved, but bitstream volume increases
Solution Approach 1:
The patent applies partial action by encoding neural-network post-filter activation information only when necessary - specifically when a filter is actually applied. The SEI messages are generated conditionally based on filter usage, avoiding redundant encoding of activation information for cases where no filter is applied, thus maintaining reliability while minimizing bitstream volume increase.
Data Source
AI summary
An image encoding/decoding method, a bitstream transmission method, and a computer-readable recording medium on which a bitstream is stored are provided. The image decoding method according to the present disclosure comprises the steps of acquiring a neural-network post-filter (NNPF) supplemental enhancement information (SEI) message; determining, on the basis of at least one neural-network post-filter characteristics (NNPFC) SEI message included in the NNPF SEI message, at least one neural network that can be used as a post-processing filter, on the basis that the NNPF SEI message is applied to a current picture; and determining, on the basis of at least one neural-network post-filter activation (NNPFA) included in the NNPF SEI message, whether a target neural-network post-processing filter applicable to the current picture is activated, wherein the target neural-network post-processing filter is determined as a neural-network post-processing filter of the last NNPFC SEI message, from among NNPFC SEI messages, of a decoding order.