Image Compression Neural Network Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image compression methods struggle to significantly improve compression efficiency, often relying on adjusting internal algorithms or using neural networks to only adjust parameters or improve prediction blocks, while maintaining color information, which limits overall efficiency.
Innovation Solution
The proposed method involves converting a high resolution color image into a low resolution monochrome image using a lower resolution neural network, transmitting this image, and then restoring it to a high resolution using a super resolution neural network, while also applying a colorization neural network to restore color information, thereby improving compression and restoration efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing image compression methods maintain color information during compression, then image quality is preserved, but compression efficiency is limited
Solution Approach 1:
The patent segments the image processing into separate monochrome and color components. The compression system processes luminance (Y) and chrominance (CbCr) channels independently, allowing different compression strategies for each component. This segmentation enables more aggressive compression of the color channel while maintaining overall image quality through selective processing.
Solution Approach 2:
The patent applies parameter changes by using a neural network to dynamically adjust compression parameters based on image content. The system transforms the image to YCbCr color space and applies different quantization parameters to different channels, with the neural network optimizing these parameters to balance compression efficiency and quality preservation.
2Reliability
If neural networks are used to adjust parameters or improve prediction blocks in existing compression methods, then some performance improvement is achieved, but overall compression efficiency cannot be significantly improved
Solution Approach 1:
The patent implements a segmented approach where the neural network processes only the chrominance channel (CbCr) separately from the luminance channel (Y). This allows the neural network to focus computational resources on recovering color information that was heavily compressed, while the luminance channel uses traditional efficient compression methods.
Solution Approach 2:
The patent introduces an intermediary color restoration neural network that sits between the compressed color data and the final output. This intermediary component transforms the heavily compressed color information back to high quality, acting as a mediator that reconciles the trade-off between compression and quality.
3Quantity of substance
If image compression converts to lower resolution formats, then data transmission requirements are reduced, but image quality degradation occurs
Solution Approach 1:
The patent applies parameter changes by transforming the image to YCbCr color space and applying different sampling rates to different channels (e.g., 4:2:0 or 4:2:2 subsampling). This allows significant reduction in data size for the color channels while maintaining luminance quality, and the neural network further restores the color information to achieve final high quality output.
Solution Approach 2:
The color restoration neural network serves as an intermediary that reconstructs high-quality color information from heavily downsampled or compressed color data. This intermediary process enables the system to transmit fewer color data points while still recovering full-quality color images at the receiving end.
Data Source
AI summary
A method for restoring a compressed image according to an embodiment of the present disclosure includes receiving monochrome image data and low resolution color image data generated from an original color image of the monochrome image data, decoding the monochrome image data and generating a low resolution monochrome image, decoding the low resolution color image data generating a low resolution color image; processing the low resolution monochrome image and generating a high resolution monochrome image in accordance with a super resolution imaging neural network; and generating a high resolution color image based on the low resolution color image and the high resolution monochrome image in accordance with a colorization imaging neural network. The imaging neural network of the present disclosure may be a deep neural network generated by machine learning, and images may be input and output in the Internet of Things environment using a 5G network.