Research On Image Segmentation Based On Fuzzy Clustering For Gray Image

Image segmentation which is the key link between image processing and pattern recognition is to divide an image into several non-overlapping and meaningful areas with the same nature. Among numerous image segmentation algorithms, Fuzzy C-means converts segmentation into optimization problem, and realizes fuzzy partition of image pixels based on iterative optimization objective function according to the degree of each pixel belonged to different parts. The greatest character of this algorithm is that FCM introduces the fuzzy theory, making it retain more image information than Hard C-means. As FCM converge fast, has intuitive geometric meanings and is able to process fuzzy information, it has been widely used in many fields. However, there are some shortcomings in FCM as follow:(1) It needs to determine the classification number manually,(2) It is sensitive to initial value and easy to fall into local optimum,(3) It doesn’t consider spatial information and isn’t robust to noise,(4) It’s time-consuming.China is a big agricultural country, and the soil is the basis of agricultural development. Soil plays an important role in crop growth as well as nutrient transformation and cycles, which is vital resource for providing nutrient to crops.Nowadays, the analysis method of soil structure includes screening method, slice method and CT(computed tomography) as well as MR(magnetic resonance) method.These method have different degrees of shortcomings and are hard to be widespread used. Soil pore is the most important indicator of water flow and solute transport, and is considered to be the major object for analyzing soil image. Reasearchers can study soil structure from soil pores. Image segmentation technologies partition soil image with merit of real time and low cost. After segmenting the soil image, we can extract the pores of soil, which can found subsequent regression model to infer soil structure.This thesis proposes modified algorithms based on the shortcomings of FCM andapplys new FCM into image segmentation. The main research work can be summarized as follow:(1) Two Fuzzy C-means algorithms incorporating spatial information are proposed.The new algorithm uses WFCM to identify the cluster centers quickly, thendefines the spatial coefficient and the spatial membership to consider spatialinformation, making FCM be resistant to noise. The second improved algorithmmodifys the membership’s hard partition matrix to let FCM be insensitive tonoise.(2) Enhanced Contrast Fuzzy C-means(En CFCM) is proposed. En CFCMsubstitutes the summation of neighbor pixels’ membership for the membershipof center pixel, which improves the possibility of pixels belonged to the samecluster. En CFCM enhances the constrast of segmented image for extracting soilpores.(3) Weighted Fuzzy C-means image segmentation based on Affinity Propagation(APWFCM) uses AP(Affinity Propagation) to identify cluster number andinitial cluster centers, and then WFCM(Weighted Fuzzy C-means) partition thegray image quickly. The new algorithm resolves the problem that the clusternumber of traditional FCM needs to be determined beforehand.