Interactive Medical Image Segmentation Algorithm

Segmentation of bio-medical images has been paid much attention to in recent decades. Based on bio-medical image segmentation, the image contents can be better analyzed, which will assist doctors to diagnose correctly. Bio-medical images have the properties of low-resolution, being noisy and huge characteristic difference among various images, which all make segmentation more difficult. Therefore, segmentation of bio-medical image is of high research meaning. Since the characteristic differences among various bio-medical images are huge (so a single model cannot be suitable to different images), the thesis is focused on interactive segmentation algorithm, i.e., users first give a region of interest which includes the object to be extracted and then the region of interest is segmented.The thesis first implements the interactive thresholding segmentation algorithm and its features are analyzed. Then the thesis implements the interactive segmentation algorithm by region growing. After analyzing the features of basic region growing algorithm, a new region growing method based on edge strength is proposed. In the proposed algorithm, the region of interest is first segmented by thresholding, with which the seeds of object region are extracted. The seeds of background region are chosen as the boundary pixels along the region of interest. Secondly edge strengths are calculated by considering all the four directions, i.e., the horizontal, vertical, diagonal and anti-diagonal directions. Then based on edge strengths, region growing is respectively applied to object region and background region in a progressive mode. That is, at the beginning, only pixels with low edge strength are considered and connected to the seeds by region growing, and then step by step pixels with higher and higher edge strengths are considered and connected to the seeds to form the regions. Therefore, at first only smooth regions are grew and formed inside object or background region, then texture regions are included into the object or background region with the increasing edge strength level and finally the object or background regions meet at the strong edge pixels in the image, so the algorithm will generate good segmentation. Compared with the basic region growing algorithm, another advantage of the proposed edge strength based algorithm is that no parameter need be tuned for similarity criteria, so the proposed one is easier to use for unprofessional people such as doctors. The experiments have verified the good performance of the proposed edge strength based region growing algorithm on bio-medical image segmentation.