Image Recognition Using Wavelet PCA and 2D-DCT Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition technologies face challenges in efficiently sorting and retrieving images due to limitations in face detection, normalization, and pattern recognition, particularly in handling variations in face poses, illumination, and expression, as well as the need for extensive retraining with new image additions.
Innovation Solution
The method involves training image classifiers by identifying and normalizing face and peripheral regions within images, using a combination of wavelet domain PCA and 2D-DCT classifiers that do not require retraining for new images, and employing color correlogram and wavelet texture analysis for robustness across different conditions, along with incremental training to update feature vectors efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional face detection and normalization methods are used, then basic image recognition is achieved, but the system cannot handle variations in face poses, illumination, and expression effectively
Solution Approach 1:
The patent segments the face recognition process into multiple specialized classifiers: wavelet domain PCA for structural features, 2D-DCT for frequency domain characteristics, and color correlogram for color distribution patterns. Each classifier processes specific aspects of face appearance, enabling the system to handle variations in pose, illumination, and expression more effectively while maintaining recognition accuracy.
2Reliability
If extensive training is performed with existing image recognition technologies, then recognition accuracy is improved, but significant time and effort are required for retraining when new images are added
Solution Approach 1:
The patent performs preliminary training of multiple classifiers (wavelet PCA, 2D-DCT, color correlogram) on the complete image collection during an initial training phase. This preliminary action creates a robust feature vector database that can accommodate new images without requiring extensive retraining, as the classifiers are already configured to recognize patterns across the full range of variations in the collection.
3Adaptability or versatility
If multiple classifiers are used to improve recognition robustness, then handling of variations in face appearance is improved, but system complexity increases
Solution Approach 1:
The patent merges multiple classifiers (wavelet PCA, 2D-DCT, color correlogram) into a unified recognition system where each classifier contributes different feature representations. The results are combined through feature vector integration, allowing the system to handle variations in face appearance comprehensively while managing complexity through systematic integration rather than separate processing systems.
4Reliability
If traditional color pattern matching is used, then basic image distinction is achieved, but robustness across different conditions (illumination, pose, expression) is insufficient
Solution Approach 1:
The patent transforms color pattern matching from a single-domain approach to a multi-domain representation by extracting features in different parameter spaces: spatial domain (color correlogram), frequency domain (2D-DCT), and wavelet domain (PCA). This parameter transformation enables the system to maintain recognition consistency across varying illumination and pose conditions by capturing invariant characteristics in multiple representations.
Data Source
AI summary
A method and system for image recognition in a collection of digital images includes training image classifiers and retrieving a sub-set of images from the collection. For each image in the collection, any regions within the image that correspond to a face are identified. For each face region and any associated peripheral region, feature vectors are determined for each of the image classifiers. The feature vectors are stored in association with data relating to the associated face region. At least one reference region including a face to be recognized is/are selected from an image. At least one classifier on which said retrieval is to be based is/are selected from the image classifiers. A respective feature vector for each selected classifier is determined for the reference region. The sub-set of images is retrieved from within the image collection in accordance with the distance between the feature vectors determined for the reference region and the feature vectors for face regions of the image collection.