Image Segmentation And The Application In Medical Images Based On Mathematical Morphology

Image segmentation based on mathematical morphology is one of interesting fields of image processing recently. Image segmentation based on mathematical morphology has mainly focused on two parts: one is edge detection based on morphological erosion or morphological dilation; the other is region segmentation based on watershed transform. In the thesis the two parts are studied particularly. Firstly the general theory of mathematical morphology is introduced and then the basic principle of image segmentation is studied. At last, my research work has been done as follows:On the one hand, various morphological algorithms of edge detection and region segmentation are discussed respectively in grayscale images. In the part of edge detection, superiority of morphology in the image edge detection is studied, omni-directional multi-scale morphological structuring elements are defined, an approach of image edge detection based on omni-directional multi-scale morphology is constructed by weighted combination of morphological operation, the experimental results demonstrate that the method gains better effect aimed at noisy image and non-noisy image. In the part of region segmentation, watershed algorithm is analyzed. Image segmentation based on the graph concept is analyzed in our thesis, a boundary finding problem is formulated as a shortest-path problem in a graph, then a watershed algorithm based ort image foresting transfonn is introduced, which is a unified and efficient approach for simplifying image processing problems to a minimum-cost path forest problem . The effects of filtering and threshold-setting are discussed in the image segmentation, the results of the experiment show that this method can provide accurate and closed region contours.On the other hand, various algorithms of morphological arithmetic of binary image and grayscale image are analyzed deeply, then expanded to color image. In the edge detection of color image, color image is decomposed as three color orthogonal feature: I₁,I₂,I₃ in Karhunen-Loeve (K-L) transformation method, omni-directional morphological structuring elements are defined in order to detect the edge of images in different color space , the results of simulation demonstrate that the method performs better not only in noise-suppression but also in color variation than classical edge detection operator. Through analyzing the different visual acuity of the human eye to the three primary colors (RGB), a color image filtering method based on multi-scale morphology is proposed in this thesis, experimental results show better filter effect on removing salt-and-pepper noise comparing with other filters according to subjective and objective evaluation.The above algorithms are applied to medical image segmentation finally, the experimental results show that better and robust performance could be obtained by our designed morphological algorithms, which may prepare for image retrieval based on segmentation results.