Adaptive Image Compression for Reliable Face Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Surveillance camera systems face challenges in accurately recognizing faces due to image degradation from high compression ratios and network instability, leading to misidentification of individuals.
Innovation Solution
A method and device for processing images that adjust compression ratios based on the overlap and similarity of recognized regions of interest and feature vectors, using image compression standards like H.264/AVC and HEVC, to improve face recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a high compression ratio is used to reduce storage capacity and network bandwidth requirements, then data transmission efficiency is improved, but image quality degrades causing face recognition accuracy to deteriorate
Solution Approach 1:
The patent applies local quality by differentiating compression strategies for different regions within the image. Specifically, regions containing faces or regions of interest (ROIs) are assigned lower compression ratios to preserve recognition quality, while other regions use higher compression ratios. This is implemented by detecting face locations, determining ROIs based on these locations, and then applying adaptive compression where the compression ratio varies by region rather than uniformly across the entire image, thus resolving the contradiction between overall compression efficiency and local recognition accuracy.
2Reliability
If network bandwidth is limited or network instability occurs, then data transmission reliability deteriorates, but using lower compression ratios increases data volume and exacerbates bandwidth constraints
Solution Approach 1:
The patent employs parameter changes by dynamically adjusting the compression ratio parameter based on network conditions and detected face/ROI locations. Rather than using a fixed compression ratio, the system modifies this parameter adaptively - lowering compression ratios for regions containing critical recognition information (faces/ROIs) while maintaining higher ratios elsewhere. This allows the system to optimize the balance between data volume and transmission reliability, ensuring that essential recognition data is preserved even under bandwidth constraints.
3Ease of operation
If compression encoding and decoding operations are performed, then data transmission is enabled, but encoding errors and packet loss cause image frame degradation
Solution Approach 1:
The patent applies preliminary action by performing face detection and ROI determination before the compression encoding process. By identifying which regions contain faces and should be prioritized for quality preservation, the system can pre-assign lower compression ratios to these critical regions before encoding occurs. This preliminary identification and classification of important regions allows the encoding process to focus resources on preserving recognition-critical areas, thereby mitigating the quality degradation that would otherwise result from uniform high-compression encoding and potential packet loss.
Data Source
AI summary
A processing device and an operating method of the processing device are provided. An embodiment of the present disclosure provides the processing device that obtains a reconstructed image of the raw image by compressing a raw image at a preset compression ratio, encoding the compressed raw image, and decoding the encoded image, recognizes, in the raw image, a target object and a first region of interest (ROI) including the target object and extracts a first feature vector from the target object recognized in the raw image, recognizes, in the reconstructed image, the target object and a second ROI including the target object and extracts a second feature vector from the target object recognized in the reconstructed image, determines whether the target object is recognized in the reconstructed image, based on a degree of overlap between the first and second ROIs and a degree of similarity between the first and second feature vectors, and adjusts the preset compression ratio according to whether the target object is recognized.