Research On Theory And Method Of Image Segmentation Based On Synthetic Feature

Image segmentation is a classic inverse problem which consists of achieving a compact region-based description of the image scene by decomposing it into meaningful or spatially coherent regions sharing similar attributes. It is a first task of image analysis process. The subsequent tasks of image analysis rely heavily on the quality of the segmentation. In recent years with the development of technologies, such as machine vision, pattern recognition and content-based image retrieve etc, and the wide use of color images, image segmentation, especially color image segmentation has emerged to be one of the hot research areas in image domain.In this dissertation, lots of research work has been done around some deficiencies of Fuzzy C-Means Clustering(FCM) in image segmentation and Color- and Texture-based Image Segmentation. The main contributions of this dissertation are summarized as follows:1. Fuzzy c-means (FCM) clustering is one of well-known unsupervised clustering techniques, which has been widely used in automated image segmentation. However, when the classical FCM algorithm is used for image segmentation, there are also some problems, such as the heavy calculating burden. We present an efficient algorithm to implement a FCM clustering that produces clusters comparable to slower methods. In our algorithm, we partition the original image dataset into unit blocks, mark the blocks by area quantization and cluster the centroids of the unit blocks. In this way, we can dramatically reduce the time for calculating final converged centroids. Experiments show that this algorithm produces comparable clustering results as FCM algorithm, but has much faster computation speed than classical FCM algorithm.2. Color image can provide more perceptual information, color image segmentation is being paid more and more attention. We propose a new approach for color image segmentation that is based on low-level features for color and texture. It is aimed at segmentation of natural scenes, in which the color and texture of each segment does not typically exhibit uniform statistical characteristics. Firstly, the local color composition in terms of spatially adaptive dominant colors is described by using Gibbs random filed, and the color image is segmented into regions according to the local color composition. Secondly, the texture characteristics of the grayscale component are described by utilizing the Steerable filter, and the grayscale component of color image is cut into flat regions and no-flat regions. Thirdly, the local color composition and texture characteristics are combined to obtain an overall crude segmentation. Finally, an elaborate border refinement procedure is used to obtain accurate and precise border localization by appropriately combining color-texture features with the Normalized cuts. The experimental results demonstrate that the color image segmentation results of the proposed approach hold favorable consistency in terms of human perception.