Face Recognition Dictionary Creation via Local Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition techniques face challenges in achieving high accuracy when face orientation, expression, and illumination conditions vary, leading to increased registration and processing requirements due to the need for multiple face images under different conditions.
Innovation Solution
An image recognition apparatus and method that detects local regions from images, calculates feature amounts, and registers dictionary data based on similarity, allowing for efficient recognition with reduced data requirements by selecting necessary local features and registering only those with high similarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple face images under various conditions (orientation, expression, illumination) are registered to improve recognition accuracy, then recognition accuracy is improved, but the number of registered images and processing amount increase
Solution Approach 1:
The patent divides the face image into multiple local regions (e.g., left eye, right eye, nose, mouth) and extracts features from each region separately. This segmentation allows the system to capture distinctive local characteristics without needing to register complete face images under all possible conditions, thereby reducing the quantity of registered data while maintaining recognition accuracy.
Solution Approach 2:
The patent extracts key local features from face images (such as eye shapes, nose structures, mouth contours) and uses only these extracted features for recognition. By taking out only the essential local characteristics rather than using complete face images, the system reduces the amount of data needed for registration while preserving recognition accuracy.
2Reliability
If multiple face images under various conditions are registered to improve recognition accuracy, then recognition accuracy is improved, but processing amount and time increase
Solution Approach 1:
By segmenting the face into local regions and extracting features from each region, the patent reduces the amount of data that needs to be processed during recognition. Instead of comparing complete face images under multiple conditions, the system only needs to compare extracted local features, significantly reducing processing time while maintaining accuracy.
Solution Approach 2:
The patent uses partial action by extracting only the necessary local features from face images rather than processing all possible facial characteristics and conditions. This selective extraction reduces the processing burden while providing sufficient information for accurate recognition.
3Reliability
If face recognition is performed with restricted capturing conditions to ensure high accuracy, then recognition accuracy is improved, but adaptability to various real-world conditions decreases
Solution Approach 1:
The patent applies local quality by focusing on extracting distinctive local features from specific regions of the face (eyes, nose, mouth) that are more stable and recognizable across different conditions. By emphasizing these local characteristics, the system achieves both high accuracy under controlled conditions and adaptability to various real-world conditions, as local features remain consistent despite changes in overall facial orientation, expression, or illumination.
Data Source
AI summary
An image recognition apparatus comprising: an obtaining unit configured to obtain one or more images; a detection unit configured to detect a target object image from each of one or more images; a cutting unit configured to cut out one or more local regions from the target object image; a feature amount calculation unit configured to calculate a feature amount from each of one or more local regions to recognize the target object; a similarity calculation unit configured to calculate, for each of one or more local regions, a similarity between the feature amounts; and a registration unit configured to, if there is a pair of feature amounts whose similarity is not less than a threshold, register, for each of one or more regions, one of the feature amounts as dictionary data for the target object.