A new hybrid face recognition algorithm based on discrete wavelet transform and direct LDA

• Published in: 2016 23rd Iranian Conference on Biomedical Engineering and 2016 1st International Iranian Conference on Biomedical Engineering (ICBME) pp. 267 - 270
• Main Authors: Bagherzadeh, Seyyed Amir Ziafati, Sarcheshmeh, Alireza Noei, Bagherzadeh, Seyyed Hassan Ziafati, Khalilzadeh, Mohammad Mahdi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 2016
• Subjects: Classification algorithms; dimensionality reduction; direct LDA; discrete wavelet transform; Discrete wavelet transforms; Face; Face recognition; feature extraction; Kernel; Principal component analysis; support vector machine; Support vector machines
• DOI: 10.1109/ICBME.2016.7890969

Face recognition (FR) has received significant attention as one of the most successful applications of image analysis and understanding, during the past several years and is an active yet challenging topic in computer vision applications. Also potentially will help in identifying ultra-rare and developmental disorders. Linear discriminant analysis (LDA) has been widely used for feature extraction in face recognition. However, the main deficiency of LDA-based algorithms is small sample size (SSS) problem, which makes between-class scatter matrix incomputable. Since the zero eigenvalues in within-class scatter matrix contains significant discriminatory information, a direct LDA (DLDA) based algorithm is proposed to save the possible useful information. In this paper, a new hybrid FR algorithm proposed using discrete wavelet transform (DWT) with third-level of Haar filter, DLDA method for dimensionality reduction, and a support vector machine with second-order polynomial kernel. The obtained recognition results show that this approach significantly outperforms recognition using proposed algorithm. For the ORL face database, the averaged recognition accuracy is over 95%.