Face Detection Video Encoder Bitrate Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional videoconferencing systems do not optimize compression methods for focused regions of interest, such as faces, leading to low video quality during movement due to blocky or pixelated images and compression-related deficiencies.
Innovation Solution
A video encoder pipeline with a face detection module that allocates more bits to face data based on activity levels and detection certainty, using a combinatorial logic algorithm to prioritize face data encoding, enhancing bitrate allocation and quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If uniform compression is applied to all areas of the video scene, then the overall bitrate is reduced, but the quality of focused regions (faces) deteriorates during movement
Solution Approach 1:
The patent applies different compression qualities to different regions of the video frame based on their importance. Face regions are identified through detection algorithms and allocated higher quality compression parameters (lower quantization values) compared to background regions. This allows the system to maintain high quality where needed (faces) while using more aggressive compression in less important areas, resolving the contradiction between overall bitrate reduction and focused region quality.
2Manufacturing precision
If more bits are allocated to face data, then the quality of facial features is improved, but the overall bitrate increases
Solution Approach 1:
The system implements region-specific bitrate allocation where face regions receive preferential bit allocation. The video encoder divides the frame into important regions (faces) and less important regions (background), applying different quality targets to each. This ensures that facial features maintain high quality with sufficient bit allocation while the overall bitrate is controlled by applying more aggressive compression to non-face areas.
Solution Approach 2:
The video frame is segmented into multiple regions based on their semantic importance. Face detection algorithms identify and segment facial regions from the rest of the scene. This segmentation enables the encoder to apply different compression strategies to different segments, allocating more bits to face segments while maintaining lower bitrates for background segments, thus improving facial quality without proportionally increasing overall bitrate.
3Productivity
If compression is increased to reduce bitrate, then transmission efficiency is improved, but compression artifacts (blocking, ringing) increase in quality-critical areas
Solution Approach 1:
The patent implements spatially-varying compression quality where face regions are protected from aggressive compression that causes artifacts. By identifying face regions and applying lower compression ratios (higher quality settings) specifically to these areas, the system maintains transmission efficiency through overall compression while preventing blocking and ringing artifacts in quality-critical facial regions.
Data Source
AI summary
Embodiments include a codec for use in a videoconferencing or similar system includes a video encoder pipeline that has a pre-processor component that is optimized to detect faces and compress the facial video data in an optimum manner. The codec has a pre-processing step that analyzes each frame on a per macroblock basis to determine the mathematical activity level per block. The activity level calculation is used as a parameter to the bitrate control module of the encoder to control the quantization, and thus the fine grained quality of the output data. An object detection module (e.g., a face detector) is placed in the pre-processing step. The object detection data is then combined with the activity level and object detection certainty value through a combinatorial algorithm comprising a weighted average or normalized multiplication process.