Adaptive Image Coding with Neural Network Artifact Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image coding techniques face challenges in efficiently compressing image data without introducing perceptible artifacts, particularly due to improper selection of pixel block sizes and data loss during quantization, which affects the quality of reconstructed images.
Innovation Solution
The proposed solution involves adaptive coding techniques that partition image data into pixel blocks and utilize multiple coder systems operating at different sizes, with downsampling and upsampling to match coding sizes, combined with neural network processing for quality enhancement and artifact minimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If pixel blocks are coded at reduced size through downsampling, then compression efficiency is improved, but image quality deteriorates due to quantization losses and artifacts
Solution Approach 1:
The image is divided into multiple pixel blocks that are processed independently at different downscaling levels. Each pixel block can be coded at its original size or downsampled based on local characteristics, allowing selective compression without uniformly degrading entire images. This segmentation enables different regions to have different compression ratios based on their importance and content complexity.
Solution Approach 2:
Different pixel blocks are assigned different coding qualities based on their local characteristics. Regions with important visual information (edges, textures, faces) are coded at higher quality with less downsampling, while homogeneous regions are coded at lower quality with more aggressive downscaling. This local quality approach optimizes the trade-off between compression efficiency and perceived image quality.
2Loss of energy
If pixel block sizes are increased for better compression, then bandwidth usage is reduced, but artifact perception increases due to improper block size selection
Solution Approach 1:
The pixel block size and downscaling level are dynamically adjusted based on the content characteristics of each region. Rather than using a fixed block size for the entire image, the system adapts the coding parameters locally, selecting appropriate block sizes and downscaling factors that minimize artifacts while achieving compression. This dynamic adaptation prevents the visibility of block-based artifacts that occur with improper fixed-size partitioning.
Solution Approach 2:
The coding parameters including block size, downscaling factor, and quantization strength are changed based on local image characteristics. Regions with high frequency content use different parameters than regions with low frequency content, optimizing the balance between compression and artifact reduction. This parameter adaptation allows the system to achieve better compression without introducing perceptible artifacts.
3Productivity
If multiple coder systems operate at different sizes with downsampling, then compression efficiency improves, but system complexity increases
Solution Approach 1:
A single coding system is designed to perform multiple functions by incorporating variable downscaling capabilities. Rather than requiring separate dedicated coder systems for each block size, the universal coder can operate at multiple resolution levels and adapt its downscaling factor based on input requirements. This multi-functionality reduces system complexity while maintaining the compression efficiency benefits of adaptive downscaling.
Data Source
AI summary
Techniques are disclosed for coding image data adaptively at different levels of downscaling. Such techniques may involve partitioning input data into pixel blocks for coding and performing content analysis on the pixel blocks. The pixel blocks may be input to block coders that operate at different pixel block sizes, which may code the pixel blocks input to them at their respective sizes. Except when a block coder operates at the partitioning size, block coders that operate at different pixel block sizes may perform downscaling of the pixel blocks to match their size with the block coders' respective coding size. A block decoder may invert the coding operations performed by the block coders, decoding coded image data at respective pixel block sizes, then upscaling decoded image data obtained therefrom to a common pixel block size. Image reconstruction may synthesize a resultant image from the decode pixel block data output by the decoders. Quality enhancements may be added by neural network processing systems that have been trained to other segmented images.