Image segmentation using edge detection and region distribution

In this paper, we propose a new concept to integrate the conventional image segmentation techniques in order to accomplish the reasonable segmentation results. First, we develop an automatic seed selection algorithm using histogram for both scale and color vector. And the luminance and chrominance are utilized in the image as a guidance to optimize the region growing and region merging. Then we explore the multi-threshold concept to generate plentiful local entropies for reasonable edge detection. Finally, for texture regions elimination, the region distribution and the global edge information are employed to identify the region with texture characterization to obtain segmentation results. In the experiment, our new technique will show more accuracy of segmentation and region classification than proposed techniques.

I. Introduction

Image segmentation is to segment an image into several regions which inner pixels are with a same homogeneous feature. The features of still image in spatial domain can be categorized into three types: intensity, colors distribution and texture. In general, the dominant criterion in image segmentation is described as follows [1]. 1.

Each segmented region must be with a single identity and represents a same characterization.

2.

The original image is composed of the segmented regions.

3.

The structure of segmented region must be possibly simplified to avoid over-segmentation.

4.

There must be significant different characterizations during the adjacent regions.

5.

The edge of segmented region must be clear and easily recognized.