Convolutional Neural Network Image Encoding for High Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing quality of digital images and videos results in larger data sizes, making it challenging to satisfy bandwidth requirements for media data traffic, necessitating more effective media data compression solutions.
Innovation Solution
An image encoding apparatus utilizing convolutional neural networks to update features of multiple images, superpose them with a primary image, generate prediction images, and determine difference features, along with an image decoding apparatus that reverses this process, enabling efficient compression and decompression while maintaining image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If image quality is improved with higher standards such as BT.2020, then image quality is improved, but data size increases significantly
Solution Approach 1:
The patent segments the image processing task into multiple parallel processing paths, each handling different image data. By dividing the overall compression task into separate convolutional neural network processing streams, the system can process multiple images simultaneously, improving compression efficiency while maintaining high image quality standards like BT.2020
Solution Approach 2:
The patent transforms the compression problem from traditional spatial domain processing to feature space processing using convolutional neural networks. By extracting and processing image features in a higher-dimensional feature space rather than directly compressing pixel data, the system achieves better compression ratios while preserving image quality
2Manufacturing precision
If media data traffic increases to support higher quality images, then image quality is improved, but bandwidth requirements cannot be satisfied
Solution Approach 1:
The patent extracts essential image features using convolutional neural networks before compression, separating the most important visual information from redundant data. By taking out only the critical features needed for high-quality reconstruction, the system reduces data traffic requirements while maintaining image quality, making it feasible to transmit within existing bandwidth constraints
3Productivity
If traditional compression methods are used, then bandwidth usage is reduced, but image quality deteriorates
Solution Approach 1:
The patent replaces traditional mechanical compression algorithms with convolutional neural network-based compression. Instead of using fixed mathematical transforms and hand-crafted compression algorithms, the system uses learned feature representations from deep neural networks, achieving superior compression ratios while preserving image quality that traditional methods cannot achieve
Data Source
AI summary
Provided are an image encoding method, an image decoding method and an image processing system including image encoding/decoding apparatus. The image encoding method includes steps of: acquiring a first image and a plurality of second images; updating features of each second image of the plurality of second images to obtain corresponding update features; superposing the first image with the update features of each second image of the plurality of second images to generate superposed images; generating a plurality of prediction images according to the superposed images; determining difference features between each second image of the plurality of second images and a corresponding prediction image; outputting the superposed images and the difference features; wherein the updating and/or predicting adopts a convolutional neural network.