A Study Of The Content-based Face Image Retrieval Technology

Nowadays, face image retrieval technology is becoming one of the popular researches in the area of face image processing. This thesis is a study on content-based face image retrieval technology, and its core analysis is about face image detection, discriminant features extraction, and similarity match. The purpose of the study is to design a content-based face image retrieval system that can retrieve the face image needed by users quickly and accurately while maintains the stability and security.Based on the above-mentioned goals, this thesis analyzes the face image retrieval technology used for generating a database by filtering image noises, discriminant features extraction and similarity match employed for detections. The followings are the main contents of the study.(1) Through the analysis of the content-based image retrieval model, this thesis proposes a face image retrieval model, and analyzes the relevant face image database with algorithms.(2) Generating a relatively standardized face image library through the filtrations of the primary data plays a great role in improving the accurate rate of the face image detection. The analysis of the existing face image detection approaches leads to a conclusion, AdaBoost algorithm is of high accurate rate and instantaneity, and it has a significant comprehensive advantage. So, this system employs the AdaBoost algorithm.(3) According to the characteristics of the face image retrieval, this thesis proposes a corresponding face recognition method named subspace method. This method’s feature extraction algorithm includes three main algorithms—principal component analysis (PCA), linear discriminant analysis(LDA), and independent component analysis(ICA). This thesis puts emphasis on the principal component analysis and linear discriminant analysis. Through the detailed analysis of the two algorithms, this thesis points out their disadvantages, and then puts forward a modified algorithm named P-LDA. At last, the three algorithms are tested by the ORL face database, and the results are compared and analyzed.(4) In accordance with the content-based face image retrieval model, meanwhile, combining the analysis, design and implementation of software engineering, this thesis completes the demonstration system of the face image retrieval. And the demonstration reveals that the system has good practicality.