The Application Of The Fuzzy Clustering Algorithm In The Image Segmentation

Image segmentation refers to a technology and process that dividing images into a number of regions which have unique property and extracting the area of interest, is the key step of completing complex processing tasks such as image recognition, target tracking.In recent years, the fuzzy c-means (FCM) clustering algorithm was known as an unsupervised clustering algorithm, is applied successfully in the field of data classification and image segmentation. The algorithm introduces the fuccy concept to the membership degree of image pixels, so that making the FCM algorithm keep much more original image information than the traditional k-means algorithm, thus receiving the widespread attention.It is first put forward by Dunn, and Bezdek popularized it.However, the traditional FCM clustering algorithm in image segmentation needed to determine the clustering number, the clustering center and the initial membership degree matrix in advance, easy to fall into local extremum, without considering spatial information, so that causing sensitive to noise.This article considers the fuzzy c-means clustering image segmentation algorithm with the spatial information as the important research object. Analyzing and comparing the merit and demerit of several classic improved FCM algorithm with spatial information, and on the basis of the neighborhood weighted FCM algorithm(NWFCM), putting forward two improved neighborhood weighted FCM algorithms. NWFCM is weak to the anti-jamming of noise, and the computation complexity of similarity measure is high, so proposed the DCT subspace neighborhood weighted fuzzy c-means clustering method. Firstly, the method combined with block thought, using discrete cosine transform (DCT) to the image patches, establishing a model of similarity measure based on local information of image block.And then define the Euclidean distance of the objective function as the weighted neighborhood distance.Finally, the method was applied to synthetic images with noise, natural images and MR images. The experimental results show that the method has a good segmentation, with strong noise resistance, computational complexity is reduced at the same time.The neighborhood weighted fuzzy c-means clustering algorithm builds the neighborhood of weight function which has a low attention to the center pixels of the image block in the image segmentation, leading to a sensitive to strong noise and handling rough to the edge texture information, so this paper proposes a FCM algorithm with the combination of wavelet transform and improve neighborhood weights. The algorithm firstly handling the noise with adaptive threshold of wavelet multi-resolution analysis on the basis of the original gray image. And establishing neighborhood weight function based onlocal spatial neighborhood information and grayscale range information of image blocks on the reconstructed image with the thought of the combination of bilateral filtering. The experimental results show that the proposed algorithm has a higher degree of precision segmentation than the traditional FCM algorithm and improved FCM algorithm, and more robustness to the strong noise, image edges are more smooth.