An Efficient Face Detection in Multi-Face Images Using Support Vector Machine

2 Abstract: This paper presents a robust algorithm for face detection in still gray level images. The structure and characteristics of the human nose is used to find possible face regions. Line detection filters are employed for this purpose; furthermore from among the several candidates detected in an image, a trained Support Vector Machine is used to correctly identify a human face. The proposed method is robust to deal with illumination problems. The accuracy of this method is higher than 90%, if tested for less than 10 faces in a simple background with adequate illumination. Owing to its simplicity it can be transferred from a PC to embedded device, making it a potential for customized and miniature systems. There are many unwanted elements in a picture which are commonly known as noise and should be removed from an image for further processing. Median filter is normally used to reduce noise in an image and for preserving useful details in the image. Adaptive filtering is more selective which helps for preserving edges and other high frequency parts of an image. Adaptive median filter is applied on the noisy image and again passed through SVM.