Study Of Image Segmentation Methods Based On Region

Image segmentation is a process and method which divides original image into disjoint regions with special properties, and extract the target of interest based on discontinuity and similarity. Image segmentation methods based on regions are studied in this paper, and they divides image focused on the similarity of regional characteristics. This paper focuses on two image segmentation algorithms based on region: Otsu algorithm and watershed algorithm. Otsu algorithm uses one or more thresholds to divide image into two or more regions based on the histogram. Watershed algorithm divide image into disjoint regions with specific characteristics through the "immersion" process. This thesis gives the following content:(1) An improved 2D Otsu algorithm based on the average variance and the neighborhood information is proposed to overcome neglecting edge information and long time computing. In order to improve the accuracy of segmentation and reduce the running time, 2D histogram is reconstructed with the median filter, and then a new 2D threshold selection function is defined with the average variance.(2) A new watershed algorithm based on multi-scale and self-adaptive marker-extraction is proposed to overcome over-segmentation. The V component of HSV space represents the brightness, nothing to do with the color,but the H component and the S component are concerned with the color information. Filtering is executed on the V component, so that not only the noise can be eliminated, important color information can be retained.(3) A new watershed algorithm based on improved gradient and self-adaptive marker-extraction is proposed to overcome over-segmentation. In order to improve over-segmentation, two groups of structural elements are used to calculate gradient. And then, the markers of reconstructed gradient are extracted by Otsu algorithm.