Research On Improved Constrained Fuzzy Clustering Algorithm For Image Segmentation

Image segmentation is one of the key steps from image processing to image analysis, which is mainly used in medical image processing, remote sensing image processing, pattern recognition and computer vision. In recent years, image segmentation has caused widespread concern of many scholars. With the study of image segmentation technology, scholars have put forward a number of segmentation methods, in which the fuzzy clustering is one of the most important methods of extensive research and application. In the existing fuzzy clustering algorithm, fuzzy C-means algorithm (FCM) is the most widely used and more successful. However, FCM algorithm is sensitive to noise and outliers, without considering the image spatial information. In the process of dealing with the noise image, we often can't get the ideal segmentation results. In order to improve these defects, scholars have put forward a number of improved FCM algorithms. Fuzzy C-means with local information algorithm (FLICM) proposed a fuzzy factor that contains the local spatial information and local gray information to improve the robustness of the algorithm and noise-robust. To further improve the segmentation performance of FLICM algorithm, it was extended to reproducing kernel space and utilizing pixel neighborhood variance information update fuzzy weight factor to obtain fuzzy C-Means clustering with local information and Kernel metric algorithm (KWFLICM) with strong robustness and noise-robust performance. But for the low-contrast gray image, KWFLICM algorithm's the segmentation and noise-robust performance was relatively poor. In the fuzzy C-means algorithm with the virtual focus (FCMFP), the observation point was introduced into the clustering objective function of the FCM algorithm to obtain form a regular term. Regular factor controlled the location of the observation point (the observation point was equivalent to the focus of optical lens), and we could get different segmentation results and improve the robustness of the algorithm by adjusting the regularization factor.Therefore, this paper studied fuzzy C-mean algorithm that was more popular in fuzzy clustering algorithm and kernel spatial fuzzy local information C-means algorithm, and improved the deficiency of FCM algorithm and KWFLICM algorithm. This paper mainly completed the following work:1. A fuzzy clustering algorithm for neighborhood information of the kernel space based on noise distance was proposed to tackle the problem that fuzzy C-Means clustering with local information and Kernel metric algorithm had poor noise-robust for low contrast image. On the basis of the existing KWFLICM algorithm, this algorithm changed constraints on membership and introduced noise distance ? and got a improved noise-robust clustering objective function. Based on existing noise clustering idea, this algorithm constructs fuzzy clustering iterative expressions of membership and cluster centers with good noise resistance. Then it gave the corresponding clustering segmentation algorithm. Finally experimental results showed that the improved algorithm was more superiority than that of the existing KWFLICM clustering segmentation algorithm for low-contrast image by salt and pepper noise interference.2. A regularization fuzzy clustering algorithm based on the virtual focus was porposed to tackle the problem that the segmentation performance of fuzzy C-mean algorithm was poor for gray uneven image. And it was extended to the histogram to obtain its equivalence algorithm that was a fast clustering algorithm based on the histogram. Based on existing FCMFP algorithm idea, this algorithm introduced a regularization term, got a improved clustering objective function, and deducted iterative expressions of membership and cluster centers. Then it gave the corresponding clustering segmentation algorithm. Finally experimental results showed that the improved algorithm could effectively improve the segmentation performance of FCM algorithm, which could accurately extract the target of original image to improve the segmentation result.3. A regularization fuzzy clustering with virtual focus algorithm based on weighted two-dimensional histogram was proposed to tackle the problem that a regularization fuzzy clustering algorithm based on the virtual focus had poor noise-robust. This algorithm made full use of the pixel neighborhood information and considered the affect of different dimension attributes to the clustering results, which improved the robustness to noise in a certain extent. Finally experimental results showed that the improved algorithm had good noise-robust performance for the image by salt and pepper noise interference and Gaussian noise interference.