Dynamic Arithmetic Precision in Image Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image encoding technologies, such as HEVC, face challenges in maintaining image quality at high bit depths due to reduced arithmetic precision, which limits the improvement in image quality despite supporting higher bit depths.
Innovation Solution
The proposed solution involves an image encoding and decoding system that includes a prediction unit, transform and quantization unit, and arithmetic precision management to selectively adjust arithmetic precision based on bit depth, allowing for either constant precision or implementation-oriented processing, enabling flexible encoding and decoding processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If bit shift processing dependent on bit depth is used in HEVC, then implementation cost does not increase much, but arithmetic precision is reduced and image quality does not improve
Solution Approach 1:
The patent introduces dynamic arithmetic precision selection where the arithmetic precision is not fixed but can be changed based on bit depth. The system dynamically adjusts the arithmetic precision in orthogonal transform and quantization processes according to the actual bit depth of the input image, allowing high precision for low bit depth images while maintaining reasonable precision for high bit depth images, thus resolving the contradiction between implementation cost and arithmetic precision.
Solution Approach 2:
The patent changes the parameter of arithmetic precision based on bit depth conditions. By introducing bit depth-dependent arithmetic precision settings, the system transforms from a fixed precision approach to a variable precision approach, where the arithmetic precision parameter is adjusted according to the input image characteristics, thereby improving image quality without significantly increasing implementation complexity.
2Ease of operation
If arithmetic precision is reduced for high bit depth images, then implementation remains easy, but image quality improvement is limited
Solution Approach 1:
The patent applies different arithmetic precision levels to different processing stages based on local needs. Instead of uniformly reducing precision across all operations, the system selectively applies higher precision where it matters most (in orthogonal transform and quantization) while maintaining simpler operations elsewhere, thus improving image quality without proportionally increasing implementation complexity.
Solution Approach 2:
The system dynamically adjusts arithmetic precision based on the bit depth of input images. For high bit depth images, the system maintains higher arithmetic precision in critical paths to preserve image quality, while for lower bit depth images, it can use reduced precision to maintain implementation simplicity, thus adaptively balancing ease of operation and image quality.
3Measurement precision
If constant arithmetic precision is used regardless of bit depth, then image quality is maintained, but implementation complexity increases
Solution Approach 1:
The patent introduces conditional parameter changes where arithmetic precision is adjusted based on bit depth parameters. By making the arithmetic precision parameter dependent on the input image characteristics rather than fixed, the system achieves high image quality for high bit depth images while avoiding unnecessary complexity for lower bit depth images, thus resolving the contradiction between constant precision and implementation complexity.
Data Source
AI summary
An image encoding device configured to perform prediction for a received image for each block on the basis of encoded pixels to generate prediction errors, a transform and quantization unit configured to perform orthogonal transform and quantization on the prediction errors to generate quantization coefficients, a coefficient encoding unit configured to encode the quantization coefficients, an arithmetic precision information generating unit configured to generate arithmetic precision selection information representing selection of arithmetic precision of at least one of the prediction, the orthogonal transform, and the quantization, and an arithmetic precision encoding unit configured to encode the arithmetic precision selection information.