Improved Approach of Natural Image Segmentation
一种改进的自然图像分割方法

This paper presents an improved mean shift method in segmentation of natural image. Firstly, the bandwidth of color, which is used for cluster, is estimated from image data. Then the bandwidth of color will be adjusted according to the upper vision tasks, and it can be regarded as the resolution of segmentation. The density value of every pixel is estimated according to the adjusted bandwidth of color, and the local maximum value of density is searched using direct density searching method.Lastly,the final result of image segmentation can be got through a global optimization criterion for mode merging.Under different illumination conditions and complicated background, the global optimization criterion can achieve the purpose of stabilizing the result of segmentation and the number of the segmented regions.