Size-independent image segmentation by hierarchical clustering and its application for face detection

Abstract

In this paper, we introduce a technique to detect a target object quickly. Our idea is based on onservation on the clusters into which an image is divided by hierarchical k-means clustering with space feature and color feature. This clustering method has the advantage of extracting the region of an object with some varied size. We insist that our idea should lead to detect a target object quickly, because it is not necessary to search the locations containing no targets. First, we evaluate our clustering method and second, we demonstrate that our method is effective on an object detection by applying to our face delection system. We show that the detection time can be reduced by 24%. © Springer-Verlag Berlin Heidelberg 2004.