Fundamentals of Picture Segmentation

Abstract

Picture segmentation is a term which has been used with various meanings by different people. In this treatise it will refer to the operation of looking at a scene and picking up objects from the background. In such an effort we divide the picture into different parts which have some meaning for the viewer. Thus in Fig.4.1 we lump all black parts together and thus produce a segmentation of the picture into 6 regions. A similar operation can be done on Fig.4.2, although the distinction there is not as clear. Nevertheless, we group together all points which are darker than some average value. Segmentation by brightness level is not the only way to analyze a picture. Figs.4.3 and 4.4 show examples of segmentation by texture or context. An automatic picture processor which is going to simulate human response must be able to perform these operations. This can be a formidable task if we want our system to handle cases like that of Fig.4.4. It can be very easy if we limit our attention to simple examples like that of Fig.4.1, where a simple thresholding operation suffices: Points are assigned to one of two regions depending on whether or not their brightness exceeds a threshold value. Our goal in this chapter and the next two chapters is to describe segmentation techniques which are applicable to a variety of pictures. The present state of the art is such that we cannot hope to handle examples as difficult as that of Fig.4.4. On the other hand, there exist methods for handling cases more difficult than those shown in Figs. 4.1-3.