Research On The Segmentation Algorithms Of Microcapsule Microscopical Images Based On Directional Information

The study of microscopical objects is always a very necessary part in the medical field, for it is significant to understand biological process. The analysis of microscopical images is still done by manual work nowadays, that is to say, people observe objects by eyes to analyze and test them. It brings us a lot of problems, such as hard work, high misjudgment rate, low work efficiency and so on. Considering the improvements of these problems, people are more and more interested in how to take advantage of computer to recognize microcapsule images automatically. As a result, the automatic recognition based on image processing develops one of the research topics in the field.Microscopical images are obtained by some specific imaging instruments, so they have their own characteristics. The factors of complex backgrounds such as low contrast, uneven illumination, too many noises, fuzzy boundaries and so on, make the work of recognition more difficult. On the basis of the researches of common segmentation algorithms and considering the specific shape of microcapsule, the paper proposes a series of novel segmentation algorithms based on directional information.According to the roof-edge characteristics of images researched, the measures embodying the characteristics on the spatial distribution and gray distribution, that is the divergence and the contrast, are combined to detect edges of images. Good results are obtained in the practical applications. In the meantime, the directional information of edge pixels is also obtained. Take advantage of the directional information of edge pixels to extract the skeleton of edges. The experimental results can ensure the connectivity and one-pixel width of edges and preserve the endpoints of them. Take advantage of the directional information of edge pixels, the relative relations of edge curves and the distances between the endpoints to link edges. The adhesion and breaking of edges can be disconnected and connected well respectively. It helps to the analysis and recognition of images greatly. At last, the whole algorithm flow is tested on the practical software platform developed for algorithm research and it proves the feasibility, effectiveness and robustness of the segmentation algorithms based on directional information.The innovation of the paper is as follows: 1. Considering the directional characteristics of edges, combine the divergence and the contrast to define the directional information of edge pixels to reflect the characteristics of roof-edge. 2. In the process of edge detection, add the conception of membership to make the algorithm more adaptive. In addition, a de-noising method based on directional information is proposed. 3. Take the most advantage of the directional information of edge pixels to conduct edge thinning and edge linking. The segmentation algorithms take directional information as the core and integrate edge detection, edge thinning and edge linking as a whole.