Hybrid Frequency Domain based Feature Extraction methods for Human Face Recognition

With the rising demands of public safety systems, face recognition has gained extra notice for the researchers in recent times. Real-world face recognition systems require cautious balancing of two important concerns: execution Time, recognition rate. The most important methods for feature extraction and classification are Dimension Reduction-Discrete Cosine Transform (DR-DCT), Dimension Reduction-Discrete Fourier Transform (DR-DFT) and Dimension Reduction-Difference of Gaussian (DoG) along with the Extreme Learning Machine (ELM) classifiers. The feature vector of the proposed algorithm is reduced by means of subspace method Principal Component Analysis (PCA). The execution time of the proposed DR-DFT, DR-DoG and DR-DCT algorithms along with ELM classifier is less when compared to the traditional methods such as Hough transform, Radon Transform, Discrete Fourier Transform (DFT) and Discrete Cosine Transform (DCT) for ORL face database. Similarly, Hybrid methods are proposed by combining DR-DCT &DR-DFT (Hybrid Method 1), DR-DCT & DR-DoG (Hybrid Method 2) and DR-DFT & DR-DoG (Hybrid Method3) along with ELM classifier. Hybrid Method 1 attains 98.50% recognition rate with the feaure size of 60×1 for ORL dataset. It achieves optimal execution time of 0.020 sec.