Face Detection Based on Similarity-based Clustering Algorithm

Abstract

There are many methods proposed for human face detection, but the size of filtering mask for detecting face is still a difficult problem now. In this paper, an adaptive face detection method is proposed to precisely detect faces in image with a variety of face size and contaminated by the noise such as the non-face skin-color objects, arms, object’s color similar to skin-color, wearing clothes and background object, and some of these faces overlapped. The adaptive face detection method is composed by four steps. The first, based on skin-color classification in color space system, the skin-color pixel features include both its position and color information are extracted and used to classify the skin-color pixel to generate candidate skin-color region frames by a similarity-based clustering method (SCM). By considering the face regions property the region frames can be merged, partitioned and removed to get the optimum face frame. Then a frame integration algorithm is used for merging frames if they belong to the same face. Next, a frame segmentation algorithm can be used to partition different faces in the same region and generate the optimum of face boundaries. Finally, after performing the three algorithms above, the candidate face regions will be found by rejecting most framed regions if the ratio of height to weight is over than 2.3. The detection of the face regions in a color image can be effectively achieved by performing an appearance-based method with spectral histograms as representation and support vector machines (SVMs) as classifiers.