Adaptive Lifting Wavelet Transform for Image Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional lifting methods for wavelet transforms require a large number of lifting steps, leading to increased approximation errors and inefficient compression performance, especially in lossless image compression, due to rounding errors and lack of adaptability to local image properties.
Innovation Solution
A modified lifting method that reduces the number of lifting steps and introduces a fast adaptive lifting scheme for improved prediction of image edges, allowing for reduced rounding errors and local decorrelation, enabling better compression ratios and flexibility between lossy and lossless compression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of lifting steps is increased to improve transform performance, then the approximation accuracy is improved, but the rounding errors accumulate and compression performance deteriorates
Solution Approach 1:
The patent changes the parameters of the lifting steps by introducing adaptive prediction orders and modified update rules. Instead of using a fixed number of lifting steps, the method dynamically adjusts the prediction order based on local image properties, thereby reducing the total number of lifting steps while maintaining or improving compression performance. This resolves the contradiction by optimizing the parameter (number of lifting steps) based on actual image characteristics.
Solution Approach 2:
The patent introduces dynamic adaptation to local image statistics, making the lifting process non-static. The prediction order and lifting steps are adjusted according to the local properties of the image data, allowing the system to automatically optimize the number of lifting steps for each region. This dynamic approach reduces cumulative rounding errors while maintaining high compression efficiency.
2Ease of manufacture
If conventional lifting methods are used with fixed prediction orders, then the implementation is simple, but the adaptability to local image properties is poor
Solution Approach 1:
The patent transforms the fixed prediction order into a dynamic, adaptive prediction order that changes based on local image properties. The method evaluates local statistics and automatically selects the optimal prediction order for each region, significantly improving adaptability while maintaining reasonable implementation complexity through systematic adaptation rules.
Solution Approach 2:
The patent applies different prediction orders and lifting strategies to different regions of the image based on their local properties. By analyzing local image characteristics (such as smoothness, edge density, and texture), the method assigns appropriate prediction orders to each region, thereby achieving local optimization without requiring complete redesign of the entire processing system.
3Measurement precision
If a large number of lifting steps are used to achieve integer-to-integer transform, then the transform is more accurate, but the computational complexity increases
Solution Approach 1:
The patent optimizes the computational complexity by changing the parameters of the lifting process, specifically the prediction order and the number of lifting steps. By adapting these parameters to local image properties, the method achieves high transform accuracy with fewer computational operations, thereby reducing overall complexity while maintaining precision.
Solution Approach 2:
The patent applies partial lifting steps only where necessary, rather than uniformly applying the maximum number of lifting steps to the entire image. By identifying regions that require fewer lifting steps (based on local statistics), the method reduces unnecessary computations while maintaining sufficient accuracy, thereby lowering overall computational complexity.
Data Source
AI summary
The integrated lifting transform provides both the lossy and lossless lifting wavelet transforms while sharing the same lifting chain for either lossy data compression or lossless data compression. The lifting steps can provide a lossless compression and a lossy compression directly from lossless compression while the integer-to-integer adaptive four-stage lifting wavelet transform provides the lossless compression with improved performance.