MULTIPLE-OBJECT SEGMENTATION WITHOUT SUPERVISION BY ADAPTIVE GLOBAL MAXIMUM GROUPING

In this talk, we propose a new method to segment an image into multiple objects. A multiple object segmentation problem is unstable since the result considerably depends on the number of objects given a priori. So, one of the most important tasks in solving the problem is to automatically find the number of objects. The method we proposed is not only able to find the reasonable number of distinct regions which form the image, but also performs well for noisy images. Our method is made up of two procedures. First, we developed the adaptive global maximum grouping. In this procedure, we deal with the image histogram and can automatically obtain the number of significant local maxima of the histogram. This number indicates the number of different regions in the image. Second, we derive a simple and fast calculation to segment the image composed of multiple objects. Then, we split the image into sets of pixels with similar intensity values according to the previous procedure. In the section of numerical results, we show the efficiency of our method through many experiments.