Face Recognition via Local Key Point Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional face recognition methods, such as SIFT, struggle with recognizing faces due to changes in facial images caused by pose and illumination, leading to poor performance, especially when feature points' positions change significantly.
Innovation Solution
A face recognition method that sets key points at predetermined positions in input and stored facial images, extracts descriptors from these key points, and matches them by comparing distances between corresponding key points in adjacent regions, allocating an identification symbol based on frequent allocations to improve recognition accuracy even under changes in pose or illumination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If SIFT method is used for face recognition, then object recognition performance is excellent, but face recognition performance is poor due to inability to handle pose and illumination changes
Solution Approach 1:
The patent divides the face image into multiple local regions (e.g., left eye region, right eye region, nose region, mouth region) and extracts features independently from each region. This segmentation allows the system to handle local variations in pose and illumination separately, improving overall robustness while maintaining the reliability of feature extraction.
2Ease of manufacture
If distance-based matching technique is used, then simple comparison is possible, but performance deteriorates under strong illumination or unusual pose
Solution Approach 1:
The patent applies different processing strategies to different local regions of the face. For example, regions affected by illumination changes (like under the eyebrows or around the mouth) are processed with illumination-invariant features, while regions affected by pose changes are processed with pose-robust descriptors. This local quality approach maintains simplicity in the overall matching process while improving reliability under varying conditions.
3Measurement precision
If feature points are extracted from local regions, then detection of facial image changes is improved, but position changes of feature points due to pose variations cause recognition failures
Solution Approach 1:
The patent extracts features not only from the spatial position of key points but also from the surrounding regional context. By considering the spatial relationship between multiple key points and their surrounding regions, the system can detect pose changes and compensate for position variations. This dimensional extension from point-based to region-based features improves adaptability while maintaining measurement precision.
Data Source
AI summary
The present disclosure relates to a face recognition method, an apparatus, and a computer-readable recording medium for executing the method. According to some aspects of the present disclosure, the face recognition method includes: (a) a key point setting step of setting key points at designated positions on an input face image; (b) a key point descriptor extracting step of extracting each descriptor for each key point; and (c) a matching step of determining whether the input face image matches pre-stored face images using descriptors for key points within a designated region including each descriptor for each first key point obtained from the input face image, and second key points of pre-stored face images which correspond to first key points obtained from the input face image.