Performance Comparison of 2D-Discrete Cosine Transform and 2D-Discrete Wavelet Transform for Neural Network-Based Face Detection

Human brain can detect faces from the images constructed in their eyes. The face detection is a computerize method of locating the face in the digital image. It is an important challenge to locate faces from uncontrolled and indistinguishable background of the digital image. This paper presents human face detection from the colored images. Skin color segmentation is used for localizations of skin colored components in the digital image. The features are extracted by using 2D-Discrete Cosine Transform (DCT2) and 2D-Discrete Wavelet Transform (DWT2). The Back Propagation neural Network (BPN) is used for training and testing purposes. In this research, three sets of images: 50, 100 and 180 have been used for experimentation. About 60 % of the images were used in training phase and rests were used for testing purpose. This paper presents the performance comparison of DCT2 and DWT2 used for feature extraction. The best detection rate of 84.03 % with the false positive rate of 5.05% was obtained using DCT2 as compare to DWT2 which provided detection rate of 83.61 % with the false positive rate of 8.43%.