Research On Segmentation Of The Grayscale Images Of Neuro Stem Cell

With the rapid development of biomedicine, the research on the neuron stem cells has become more and more important. At present, to obtain the rule how neuron stem cells develop into different nerve-cells, it is necessary to track the cleavage and prolification process of artificial cultured neuron stem cells. And the solutions always combine analysis technique of computing digital image with high speed processing in signal and information together. Against a series of problems in segmentation results lead by existing segmentation methods, new segmentation methods are employed for grayscale images of neuron stem cells. Pre-processing and post-processing of segmentation are also involved. With the images of neuron stem cells collected by Karin Althoff and Johan Degeman in Chalmers University of technology in Sweden, good segmentation results have been reached. The main research and innovation contents are stated as following:Firstly, much noise exists in the image with the effect of the imaging circumstance, which greatly affects the image analysis, to reduce this effect, this paper analyses the usual methods, such as linear filter, sharpen enhancement, morphological filtering, and enhancements based on histogram equalization. Aiming at the grayscale images in this project, this paper brings up the edge enhancement based on reconstruction filter and improved histogram equalization; at last, in order to reduce the effect of the board of culture to segmentation, mean-shift filtering is used to smooth the background. The noise reduction and enhancement methods mentioned above are applied in the pre-processing and the results show that the noise is greatly decreased.Secondly, in order to obtain full automatic segmentation method, Mean-Shift algorithm is employed as a kind of nonparametric clustering algorithm. Based on the characteristics of the grayscale images of neuron stem cells, this paper brings up the new segmentation method combined Mean-Shift and region consolidation. First, Mean-Shift algorithm is used to obtain initial full over-segmentation images, then initial regions are merged with three layers little regions merging method, which is based on distance and area information. At last, nonconvex pattern is used to correct the losing part of cells until the final segmentation results are reached. The correct number of cells can be reached and the segmentation contour approaches the original cell contour.Thirdly, in order to simplify the algorithm above, an improved geometric active contour is adopted for segmentation. In this method, the parameter of curvature is included to improve the shape change in evolution of curve for approaching the real cell contour, and the parameter of external force, which leads to the curve shape change, is adaptive. This change makes the curve reach the cell edge fast and makes the segmentation accurate. Good cell contour with clear edges can be reached, but for cells which have unclear edge, the performance of this method degrades.Finally, in order to analye cell contour, this paper brings up a new segmentation method combining watershed and regions merging based on gradient vector flow. Initial segmentation can obtain fine edge by watershed, but there is over-segmentation. Aiming at this question, a direction polling list is constructed based on the gradient vector flow image of the original image. The over-segmentation regions are merged to reach object region depending on the polling list. Last results can be obtained by merging the object region based on the elliptical shape of cells. This method can get real contour of cells, well it is sensitive to the parameters.In conclusion, the grayscale images are researched profoundly in this paper, and new methods in the paper include the image pre-processing as noise reduction and enhancement, as well as segmentation. The real image data proves that the noise reduction and enhancement algorithm in the paper can get better pre-processing results, and the segmentation method can fulfill the need of different circumstance, which lay a solid foundation for further work...