Research And Improvement Of Microscopic Cell Image Segmentation Algorithm Based On Template Matching

ABSTRACTMicroscopic cell image segmentation algorithm is the basic and important problem of the cell classification recognition and quantitative analysis, which plays an important role in many scientific and clinical applications. Cell image segmentation segments the cell from the background region by using an algorithm. If the target object segmented is a single cell, then the cytoplasm and nuclei can be further separated. If the object segmented are clustered cells, these overlapping cells should be separated first, and then separate the cytoplasm and nuclei. Given that cell dyeing process and smear production methods are quite different, and exists complex shapes in the cell image, many calculation methods have been developed and applied to the cell image segmentation. At present, a large number of cell image segmentation algorithms proposed have a good segmentation result when the illumination is uniform, the background is simple or there are only single cells in the image. But are not relatively good when the illumination is nonuniform, the background is complex, the boundary is blurry or there are overlapping cells in the image. So it is particularly important to research more intelligent and universal segmentation algorithm.In 2013, C. Chen proposed the microscopic cell image segmentation algorithm based on template matching, this algorithm has very good generality in the microscopic cell image segmentation. The experimental results of the algorithm show that the cell boundary gained by this algorithm is in good agreement with the actual cell boundary. This algorithm can get very good cell boundary even when the illumination is nonuniform or there are overlapping cells in the image. However, its drawback is that average segmenting time of each cell image is too long relative to other microscopic cell image segmentation algorithm, causing the algorithm execution efficiency is not high.Aiming at the problem that the execution efficiency of the microscopic cell image segmentation algorithm based on Template Matching(TM) presented by C. Chen is not high, this article has carried on a series of targeted research and improvement. The main research contents and innovation points are as follows:○1 Proposes an improved TM microscopic cell image segmentation algorithm based on template set reduced(TSR_TM). The new algorithm aiming at the problem that TM algorithm could produce more redundant template when creating template set, on the basis of deeply analyzing the reason why redundant template produced, extracted the shape features of the template set and calculated their similarity, then eliminated those template with exorbitant similarity to reduce the template set in the condition that the accuracy of image segmentation was not affected. Then reaches the goal of reducing the image segmenting time and improving the execution efficiency of the algorithm.○2 Proposes an improved TM microscopic cell image segmentation algorithm based on average template matching and the seeds’ optimal location selection(ATM_SOLS). The new algorithm reduced more template set though replacing the template sets produced by the TM algorithm by template set which the average template is accordingly adjusted in size and direction, and ensured the segmentation accuracy of the algorithm though improving the method of seeds selection in the process of image segmentation. This algorithm reduces more image segmenting time, improves the execution efficiency of the algorithm in the condition that the accuracy of image segmentation is not affected.○3 Carries on the qualitative and quantitative comparison of the above proposed two kinds of improved algorithm and TM algorithm experimented in U2 OS image set and NIH3T3 image set, and the performance of the improved algorithm is verified.