Color Image Segmentation Based on Statistics of Location and Feature Similarity

Fumihiko MORI, Hiromitsu YAMADA, Makoto MIZUNO, Naotoshi SUGANO

The process of “image segmentation and extracting remarkable regions” is an important research subject for the image understanding. However, an algorism based on the global features is hardly found. The requisite of such an image segmentation algorism is to reduce as much as possible the over segmentation and over unification. We developed an algorism using the multidimensional convex hull based on the density. In the concrete, we propose a simple and quick new algorism in which regions are expanded according to the statistics of the region such as the mean value, standard deviation, maximum value and minimum value of pixel location, brightness and color elements and the statistics are updated. We also introduced a new concept of conspicuity degree and applied it to the various 21 images to examine the effectiveness. The remarkable object regions which were extracted by the presented system, highly coincided with those which were pointed by the sixty four subjects who attended the psychological experiment.

IEICE Technical Report

The Institute of Electronics, Information and Communication Engineers （IEICE）

110

467

241

246

0385-4221