Transform Coding Randomization for Video Artifact Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video encoding methods produce blurry and inconsistent blocks in areas of high texture and motion, leading to poor video quality due to rigid threshold settings in quantization, which fail to balance detail representation with limited bit rates.
Innovation Solution
A novel method and system that dynamically modifies the quantization Dead Zone and transform coefficient levels by randomizing threshold values and re-computing high-frequency coefficients to enhance detail in blurry areas without significantly increasing bit representation, using a Dead Zone randomization mode and Level Zero control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If rigid threshold settings are used in quantization, then device complexity is reduced and processing is simplified, but video quality deteriorates with blurry and inconsistent blocks in areas of high texture and motion
Solution Approach 1:
The patent applies dynamics by making the quantization threshold adaptive rather than fixed. The threshold is dynamically adjusted based on local image characteristics (texture and motion complexity) through a measure computation process. This allows the system to automatically increase detail representation in complex areas while maintaining simplicity in smooth areas, resolving the contradiction between processing simplicity and visual quality consistency.
Solution Approach 2:
The patent changes the quantization parameter (threshold value) based on local image content. By computing a measure that reflects local complexity and using it to adjust the threshold, the system transforms a static parameter into a dynamic one that adapts to content requirements, thereby improving video quality consistency without significantly increasing overall system complexity.
2Manufacturing precision
If high-frequency coefficients are exactly reproduced, then manufacturing precision (video quality) is improved, but loss of information is reduced only at the cost of increased bit rate
Solution Approach 1:
The patent applies local quality by differentiating the treatment of high-frequency coefficients based on their spatial location and local image characteristics. Instead of uniformly preserving or discarding high-frequency information across the entire image, the system selectively preserves detail in regions of high texture and motion complexity while allowing more aggressive compression in smooth regions. This localized approach improves perceived quality without proportionally increasing bit rate.
Solution Approach 2:
The patent implements partial action by selectively reconstructing only the most important high-frequency coefficients rather than all of them. The measure computation identifies which coefficients contribute most to local detail, and the randomization process selectively preserves these while discarding less important ones, achieving good quality with reduced information loss compared to full preservation.
3Productivity
If quantization step size is increased to reduce bit rate, then productivity (compression efficiency) is improved, but manufacturing precision (video quality) deteriorates with loss of detail
Solution Approach 1:
The patent dynamically changes the effective quantization step size by adjusting the threshold based on local complexity measures. In regions requiring detail preservation, the threshold is lowered (effectively reducing quantization step size), while in smooth regions, the threshold is raised (increasing quantization step size). This adaptive parameter change allows the system to achieve better overall quality at a given bit rate compared to uniform quantization.
Solution Approach 2:
The system transitions from static uniform quantization to dynamic adaptive quantization. The threshold parameter is adjusted in real-time based on computed measures of local image complexity, allowing the quantization process to respond to content requirements and maintain detail fidelity where needed while achieving compression efficiency where possible.
Data Source
AI summary
A method and system are provided where image data is encoded in the spatial domain, and transformed in a two dimensional transform process to thereby recover a frequency domain representation of the image data. The frequency domain representation is then quantized to obtain an integer representation. The integer representation is ordered by frequency. Then, the hi-frequency coefficients are recreated and intelligently randomized at selective frequencies. This provides high quality encoded picture results, with fewer artifacts than those that would result from conventional approaches afflicted by artifacts due to loss of high frequencies.