Research On Microscopic Images Processing In Materials And Biomedicine

Processing, segmentation and feature extraction of microscopic images are very important in analyzing microstructure properties. It is often used as aids in material science and cell histology. This paper mainly focuses on microscopic images processing techniques in materials and biomedicine by combing with two projects supported by the country and Taiwan research institute respectively.Alloy microscopic images consist of white matrix phase and black precipitation phase. Traditional artificial method has some disadvantages, such as strong subjective, time-consuming etc, and present images processing technique based on simply connected region is not suitable for alloy images with texture arranged in a crisscross pattern in this article. Therefore, this paper presents novel target segmentation and feature extraction algorithms for alloy images.(1) OTSU method is applied to segment filtered images according to intra-region similar property and inter-region dissimilar property in the gray alloy images. Then, complementarity between small areas and holes is utilized to solve the problem of holes filling in complex textures.(2) A method is proposed to calculate the size of textures arranged in a crisscross pattern. The main idea of this method depends on the area of rectangle obtained by multiplying length by width. Firstly, morphological operators are adopted to extract the skeleton of images. Then a new pruning method is used to remove useless branches. The size of texture is capable of being calculated by the ratio of its area to the length of skeleton. Finally, some examples by using statistics are given which illustrate the efficiency and feasibility of the presented method.Morphology and size of cells may change abnormally in certain pathological condition, which can be regarded as a scientific basis to analyze and diagnose diseases. Aimed at the complexity and inhomogeneity of cell regions, this paper focuses on target segmentation and adherence separation algorithms.(1) An image segmentation method combining clustering and edge detection is given for the cell microscopic images. Partial cell regions are obtained by applying K-means clustering method in the balanced-processing images. Then, remaining domains are divided into subdomains with closed edges based on Canny operator. Those small subdomains are merged with initial cell regions according to some criterions. Finally, remaining background domains are departed by watershed algorithm and small areas are merged again.(2) Aimed at the conglutination in cells, an improved erosion-expansion method is introduced in this paper. Concave points detection and variable structural elements are adopted to solve the problems of irreversibility and the choice of structuring element in traditional method, respectively.