The Kernel Clustering Algorithm And Application In Image Segmentation

In digital image processing, image segmentation which occupies an important position, has obtained widespread research. In the research and application of image, interests are only focused on certain parts of the image. These parts are often called goals or prospects (other parts are called background), which generally corresponds with a specific region of the image which has a unique nature. These regions are needed to be separated and extracted to identify and analyze the target. Image segmentation refers to the technology and process in which images are divided into various regions and the interesting objectives are extracted.Image segmentation is essentially a classification process of all pixels of the image, therefore the application of the spatial clustering algorithm into image segmentation is very suitable. The application of features spatial clustering approach to image segmentation can be regard as the promotion a threshold segmentation concept. It shows the elements of the space images in accordance with the characteristics of their measured values with the corresponding feature space. It categorizes them by the feature space corresponding aggregation of different regions and then separates them, mapping back to the original image space.The C- average value cluster only has the good clustering effect to spherical or the approximate spherical sample but bad effect to the arbitrary shape bunch of distributed situation ,and it is also easy to fall into the partial minimum problem, .Meanwhile although the kernel clustering algorithm is generally suitable to the shape of the samples, it is every sensitive to noise. Based on the nuclear cluster algorithm and the fuzzy C- average value cluster algorithm, This paper introduces the fuzzy concept into the nuclear cluster and renews the parameter of the nuclear function by the fuzzy matrix, which avoids the unfairness of the cluster center in the computation of the mean value, when the nuclear function updated parameter. It also makes use of the advantage of the nuclear cluster which can suit the sample with arbitrary distributed shape to avoid the shortcomings of the fuzzy clustering which can only adapt in the ball samples. Synthesizing both of their merits, this paper proposes a new cluster algorithm: the nucleus fuzzy clustering algorithm, and applies it in the image division, which has got good division effects.The main idea of the algorithm is to introduce the fuzzy concept into the nuclear function and achieve good clustering effect by the of fuzzy the nuclear function parameter. This algorithm uses the nuclear function which fits the Mercer condition and maps the sampled data into the high Uygur space to enlarge the category difference between the samples, which can help achieve the aim of separateing the sample linearity.It also simplifies the distance computation by the use of the characters of the nuclear function. For the image segmentation ,compared with the general commonly used image threshold method,this algorithm avoids the difficulty in threshold Selection and it has the universal applicability to the image. Otherwise, it overcomes the disadvantage of threshold method which has good effects only to the images with the background which has remarkable contrasts ,but has no obvious effect to other images.