Image Segmentation Based On Morphology Watershed And Fisher Linear Discriminant Analysis

Image segmentation is one of the most important issues in the image processing; it is also the fundamental of low-level computer vision. Image segmentation is to represent an image as a set of connected regions. It is the foundation of image understanding and analysis. The segmented quality of image has directly impact on the image processing steps.The accuracy and speed of segmentation algorithms are important for real-time image processing system. To improve the accuracy of segmentation result and the speediness of segmentation algorithm, we fulfill the following work:Watershed segmentation based on mathematical morphology is a popular tool for image processing. It is easy, intuitive, rapidly computational and parallelizable. It can get a connected and closed profile. But there is much"over segmentation"in the segmentation results. In this paper, the definition and implementation of watershed algorithm is given firstly. Then, the reason of"over segmentation"is analyzed. Final, two improve algorithms are proposed.In the first method, a new watershed algorithm is proposed combined with wavelet transform and multi-scale morphology gradient. By choosing a right structure elements for pre-processing and transfer multi-scale gradient for the original image, the watershed is applied on the approximated image,"over segmentation"is dramatically reduced by morphological reconstruction. In the second method, an improved marker-controlled watershed segmentation algorithm is proposed. The algorithm modified minimum points as markers, and watershed algorithm is applied on the modified gradients by the markers. Experimental results show that the proposed algorithms both can overcomes the over-segmentation problem of watershed. At the same time, the second method can extract markers adaptively without the need for prior knowledge and overcome marker extraction difficulties.Image segmentation based on two-dimensional Fisher linear discriminant analysis method not only considers the spatial information of pixels, but also considers the between-class scatter and within-class scatter in the image simultaneously. It is suitable to segment images with low signal-to-noise ratio and fuzzy edge.To reduce the computation burden of two-dimensional Fisher linear discriminant analysis method, particle swarm optimization (PSO) is used to find the best two-dimensional threshold vector. Results show that PSO can find the best two-dimensional threshold vector in a short time. At the same time, image segmentation method based on two-dimensional bound histogram and Fisher linear discriminant analysis is proposed. First, bound set is constructed and then bound histogram is built to exclude the interferential components and reduce the search space. In the end, we propose a fast recursive algorithm and PSO for two-dimensional Fisher linear discriminant analysis, avoiding reduce computation complexity, and it is suitable for real time application.