Residual Image Compression Using Per-Pixel Noise Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current residual image compression techniques are inefficient, particularly in terms of bandwidth and storage requirements, due to limitations in existing image prediction mechanisms.
Innovation Solution
Implementing a noise model-based compression scheme that utilizes a predictor, such as an auto-encoder, to generate predicted images and residual images, allowing for more efficient compression by minimizing the difference between the acquired and predicted images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If current residual image compression techniques are used, then bandwidth and storage requirements are reduced, but compression efficiency is insufficient
Solution Approach 1:
The patent transforms the residual image from its original uniform format into a variable-precision format by applying different bit-depth reductions to different spatial regions. This parameter change allows the compression system to adaptively allocate bits based on local image characteristics and noise levels, thereby improving compression efficiency while maintaining perceptual quality in critical areas.
Solution Approach 2:
The patent divides the residual image into multiple regions and applies different compression strategies to each region based on its specific characteristics. By analyzing local noise levels and image importance, the system preserves higher precision in regions where it matters most while applying more aggressive compression in less critical areas, thus resolving the contradiction between compression efficiency and quality retention.
2Quantity of substance
If residual image compression is applied, then the size of residual images is reduced, but compression efficiency remains suboptimal
Solution Approach 1:
The patent introduces dynamic adaptation into the compression process by using a trained neural network to analyze each residual image and determine optimal compression parameters for different regions. This dynamic approach replaces static, uniform compression with adaptive, content-aware compression, significantly improving compression efficiency while achieving smaller residual image sizes.
Solution Approach 2:
The patent replaces traditional mechanical compression algorithms with a data-driven neural network-based system. The trained model learns optimal compression strategies from training data and automatically applies them to new residual images, achieving superior compression efficiency compared to conventional mechanical compression methods.
3Quantity of substance
If image prediction mechanisms are used, then bandwidth and storage are reduced, but limitations in prediction accuracy persist
Solution Approach 1:
The patent performs preliminary analysis of the residual image using a trained neural network before applying compression. This preliminary action identifies regions of high and low importance, allowing the system to pre-determine the appropriate compression level for each region. By performing this analysis beforehand, the system can maintain prediction accuracy in critical areas while achieving efficient compression overall.
Data Source
AI summary
Techniques are disclosed for performing residual image compression techniques used in conjunction with image and/or video predictors. The techniques utilize a compression scheme that implements a noise model to estimate noise values of pixels in an originally acquired image. These noise value estimates are then used to perform residual image compression more efficiently by performing a non-uniform reduction in resolution of the residual image. The resolution reduction includes dropping least significant bits (LSBs) used to encode each pixel on a pixel-by-pixel basis based upon the noise value estimates of the originally acquired image.