The Automatic Segmentation And Recognition Of Microscope Image Of Binuclear Cell

The recognition of Micronucleus Cell is rather difficult. To point out the Binuclear Cell is the key to solve the problem. Due to the characteristic of Microscope Cell Image, the algorithm is divided into three parts: the segmentation of nucleus; the segmentation of cytoplasm; and the recognition of Binuclear Cell.A novel method used for segmentation of nucleus is proposed: filter the origin image with the median filter in both horizontal and vertical direction, which has a fairly wide window size, in order to reduce the infection of the noise and impurity; in both direction, use the threshold combined with the technology of finding the peak in local area to detect the real nucleus edge, and thus minimize the non-nucleus edge; also in both direction, let the border growing respectively using the type information; thereafter, weed the short border and link the border together to segment the nucleus, exploiting the convex character and border type. For the further exploration, build the neighborhood information among segmented nuclei.Base on the work of nucleus's segmentation, drop the nucleus that has the most amount of neighbors, and process the cytoplasm segmentation. The pyramid algorithm here has three advantages: noise reduction due to the application of a low pass filter, contrast enhancement due to the low pass filter and sampling of image during construction of the pyramid, and decreased execution time due to the size reduction of the original image. Devolope a new extended Kirsch operator with four new templates and the variation of the both side of edge, thus give the ability to judge more direction and use the more local information. It is used with pyramid algorithm to detect the cytoplasm edge. And then check the direction and the nuclei area to eliminate the incorrect edge point. The shape of cell is almost ellipse, so the ellipse fitting technology is used here. Combined with Theil-Sen estimator and Repeated Median estimator, a new randomized method is proposed and has a good tradeoff between efficiency and complexity. Count the number of nucleus belonging to a segmented cell and then judge the cell whether it is a binuclear cell.At the same time introduce the new method, practical considerations and key classes and functions are also presented.