Abstract: Human face detection plays a very significant role in various biometric applications like crowd surveillance, human-computer interaction, automatic target recognition, artificial intelligence etc. Varying illumination conditions, color variance, pose variations affect face recognition performance. So, automatic facial detection and recognition is an interesting concept that has evoked considerable attention because of its applicability in various areas. Our work suggests a novel algorithm for enhancing the facial detection and recognition performance, which comprises of two major steps: first, we locate the faces and then the located faces are recognized. We have utilized multiple color space based skin color segmentation and morphological operations for facial detection that is faster and has more accuracy when compared with the other existing algorithms. First, skin regions are segmented from an image using a combination of RGB, HSV and YCgCr color models using thresholding concept. Then facial features are used to locate the human face depending on understanding of geometrical features of human face. The face recognition method contains four stages: Gabor feature extraction, dimensionality reduction by making use of PCA, selecting features using LDA, and classification using SVM. Simulation results show that, our suggested approach is sufficiently robust for achieving approximately 96% accuracy and recognizes faces with lesser misclassification compared to existing schemes.

Keywords: face detection, skin color segmentation, RGB, HSV, YCgCr, Sobel edge detector, LDA, SVM.