Research On Separation Algorithm For Overlapping Or Touching Objects

Image separation is the primary issue in the study of image analysis and pattern recognition. As one of the typical hard issues in image processing, there are still no general theories for image separation. Because of various applications, image quality, and difference in research objects themselves, it's very difficult to find a general separation method.Chromosome images analysis is a comprehensive research topic which involves image processing, pattern recognition, computer vision, and other fields. Chromosome image separation is the foundation for objects recognition and feature extraction, and that can directly affect accuracy of subsequent image processing. This paper focus on chromosome images, try to explore suitable separation algorithms for chromosomes, with their unique features.Since not all 46 chromosomes in one chromosome image touch or overlap to each other, there exist isolated chromosomes. But the separation procedure is based on touching or overlapping chromosomes, the existence of isolated chromosome would not only increase the image computing complexity, but also influence the accuracy of image processing. So it would be very necessary to extract the isolated chromosomes before separation. How to differentiate the isolated chromosomes with others constitutes the first research point in the study. This paper proposes a new method to define isolated chromosomes. Based on Euclidean Distance of objects, the concept of feature line is pointed out, and is compared with main skeleton line in order to make a match. The method to determine if the two lines can match constitutes criterion for isolated chromosomes. In this way, isolated chromosomes can be detected and extracted from the image to be separated, and provide the premise for the subsequent separation algorithm research.Because 24 types of chromosomes in one metaphase cell have unique textures and patterns respectively, to compute and extract these texture bands as important features of chromosomes, is instructively significant for chromosome image separation and object recognition. The second research point in this study to accurately extract texture bands for chromosome images. Considering the grey level variance between texture area and neighboring area, this paper attempts to get correct regional threshold, which is based on BERNSEN threshold algorithm, to separate texture area from neighboring area. Then iterative method is applied to achieve more regional thresholds and to conduct more binarization. In this way, the image can be greatly sharpened and the texture area can be highlighted. To avoid excessive binarization, the iterative procedure will stop before broken texture bands appear. Finally, emphasized texture area is completely extracted with relatively high accuracy. The third research point focuses on touching chromosomes, because most of the chromosome objects are touching to each rather than being overlapped. This paper attempts to find a suitable way for touching chromosome separation, with the combined information from both grey level and geometrical features. It applies iterative threshold increasing method and gradually separate touching objects with different touching degree. To be mentioned, this method can effectively separate heavy touching chromosomes, which are generally regarded as one of the most difficult research points in image separation research. Besides, it also modifies separation algorithm by using the width similarity of chromosomes to achieve satisfactory separation result.