Face Image Synthesis for Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition systems face challenges in accurately discriminating between face images due to variations in photographing conditions, such as face direction, illumination, and secular changes, which can lead to reduced accuracy and generation of dissimilar images when instance images do not contain sufficient similar individuals.
Innovation Solution
An image processing apparatus that stores face images under various conditions, approximates and synthesizes input face images to match a predetermined photographing condition, generating a new image that resembles the input face by combining instance images of multiple persons, thereby reducing the impact of variations and improving recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If instance images of multiple persons are combined to approximate input face images, then face recognition accuracy is improved when instance images are limited, but device complexity and processing time increase
Solution Approach 1:
The face image is divided into multiple parts (eyes, nose, mouth, etc.), and instance images are selectively combined for each part rather than processing the entire face as a whole. This segmentation approach improves accuracy by focusing on discriminative features while reducing overall processing complexity.
Solution Approach 2:
The system uses a limited number of instance images (e.g., 3-5 persons) rather than requiring extensive databases, achieving satisfactory recognition accuracy with partial data. This reduces the complexity burden while maintaining effective performance.
2Reliability
If only areas hardly influenced by face-direction variation are compared, then face-direction variation is deleted, but information of individual differences is lost
Solution Approach 1:
The face is segmented into multiple parts, allowing the system to compare both variation-resistant areas (for reliability) and variation-sensitive areas (for individual differences). This resolves the contradiction by processing different regions with appropriate weighting.
Solution Approach 2:
Different parts of the face are treated with different qualities of comparison - some parts are weighted more heavily for their robustness to variation, while others are used to capture individual characteristics. This local differentiation maintains both reliability and information retention.
Data Source
AI summary
An image processing apparatus stores, as instance images, a plurality of face images obtained by photographing respective faces of a plurality of persons on a plurality of photographing conditions, while associating the persons with the photographing conditions for each part; obtains the photographing condition of the input face image; approximates the respective parts of the input face images, by a combination of the instance images of the plurality of persons stored in association with the parts and the obtained photographing conditions; decides, for each part of the input face image, a combination corresponding to the combination in the approximation from the instance images of the plurality of persons stored in association with the part and a predetermined photographing condition; and generates an image obtained by photographing the input face image on the predetermined photographing condition, by synthesizing the obtained combination on the whole input face image.