Research On Overlapping Cell Identification And Segmentation Counting Method

In biomedical image, applied computer assistive technology on medical image processing has become main tools for clinical diagnosis and pathological analysis. In cell image owning to their own structure, multiple cells often overlap, which for cell image segmentation and the count is an important basis to observe and analyze whether there are disease. Cell image acquisition needs hardware, software, and many other aspects of work to do, so how to use the applied image processing and analysis technology to cut up overlapping cells image in the complex background, used to improve the accuracy of medical analysis and work efficiency for medical workers is the focus of thesis.The key factor distinguishing whether or not cells overlapped is the shape factor in morphological parameter which is an effective parameter in a measure of cell close roundness, when cells overlap, their boundaries due to sag will correspondingly lead to shape factor decrescent, therefore, we can use shape factor to effectively identify whether cells overlap to prepare for subsequent segmentation and count. In this paper, the method of calculation of cell perimeter has been improved to remove the tiny burrs on the edge of cells and the noise, making it more correspond with rounded outline of cells and proposed regional pixel labeling method based on boundary peeling-off, which avoids the problems that the cells in the process of tracking into dead circulation, and shortens the running time of the algorithm, simplifies the design procedure.Cell image due to differences in the collection process causes differences and unsatisfactory degradation of image quality. In order to more accurately show the outlines of cells, we remove the internal hole in image, and fill the image and then the image looks much smoother. In the application of mathematical morphology the gradient image to increase gray level transformation, meanwhile application of hat transform to enhance image contrast, through these image preprocessing technology eliminates irrelevant information of cell image, highlight the image features we are concerned to, enhance the cell's detectability and maximally simplify data, so as to improve image quality.To pretreated application of overlapping cells based on watershed segmentation algorithm of the controlled marker to separate, through gradient image reconstructed by morphology to get the local maximum of the image, and to mark local maximum of the image, at the same time using the peeling-off boundary method to change the binary imagine of marked local maximum in distance, obtain external tag collection, finally, do gradient correction to marked imagine and then do watershed segmentation on the modified gradient image, experiments prove that using the priori knowledge of tags, avoiding the problems of irregular segmentation and gradient occurred in the traditional watershed, finally, we do count statistics for the segmented cell image.