The Application Of Improved Fuzzy C Means Clustering Algorithm In Image Segmentation

Image segmentation is the technique and process of decomposing image to characteristic areas and extracting interesting targets and it’s the basis of image analysis, recognition and understanding. It’s an extremely important aspects of image processing and a difficult problem in image processing technology. In the process of the analysis and processing for image, the segmentation is its essence and determines the final image quality analysis and image understanding.People have done a lot of segmentation research, but it is still so far from the needs to solve practical problems. Since the image will be affected by projection, mixing, distortion, noise and the other factors in the imaging processes, the image features will be blurred and deformed to a certain extent, and then the image segmentation problem itself will have fuzziness, such as: 1) spectrum fuzziness, mainly for the mixed pixels and endmember spectral variability; 2) the ambiguity of the relationship between geometric space, mainly for spatial data uncertainty, cognitive uncertainty and analysis processes uncertainty, and so on. To solve these problems, in recent years some scholars introduced the fuzzy theory into the processing of image, applying the fuzzy theory to carry on the image segmentation. This paper studies Fuzzy C Means(FCM), which is the most commonly applied in the fuzzy clustering algorithm, and its improved algorithm in image segmentation.Fuzzy clustering algorithm has a strong practicality, but the traditional Fuzzy C Means(FCM) algorithm has limitation of equal partition trend for data sets, without considering the effect of clustering produced by actual distribution of the different samples. When all kinds of samples of data set have difference intensity, the clustering result is not very satisfactory. Therefore, this paper presents Fuzzy C Means algorithm based on a Density function weighted(DFCM algorithm). The algorithm considers the natural distribution of the data characteristics(Some data points around the data points more so the value of the point density is big; on the contrary, some data points around the data points less so the value of the point density is small). Through calculating the Gaussian density function values for each data object and normalizing them as weights, adding it to the traditional Fuzzy C Means algorithm. This paper we use the Fuzzy C Means clustering algorithm based on density function weighted to image segmentation, and proved DFCM clustering algorithm’s validity by four experimental groups. Four experimental groups are: the segmentation on gray images, the segmentation on noise images, the segmentation on color images and the segmentation on texture images. In the experiments, we use the Fuzzy C Means algorithm and Fuzzy C Means algorithm based on a Density function weighted to different images. The experimental results show that:1.For gray images, color images, noise images, texture images and remote sensing images, the convergence of DFCM algorithm is significantly improved than FCM clustering algorithm.2.The improved algorithm(DFCM algorithm) has no robustness to the noise.3.For texture images, when the texture structure difference is larger, the DFCM segmentation effect is better than the FCM algorithm. When the texture structure difference is small, the segmentation effect is worse than the FCM algorithm, and it needs the following improvement.