The Research Method Of The Segmentation And Recognition For Red And White Sells In Stool Microscopic Images Of Human

The research method of the segmentation and recognition for red and white cells in stool microscopic images of human involves the knowledge of many subjects such as digital image processing and image pattern recognition.Compared with the blood and urinary sediment images,the tangible objects of tool microscopic images have more complicated background and blurred boundary.At present,the analysis for stool image mainly relies on manual inspection in China.This method not only increased the workload of the medical inspectors,but also reduced the diagnostic efficiency.This paper combines the related knowledge,and probes into the preprocessing and the segmentation of stool microscopic images,the analysis and extraction of cell features,the classification and recognition algorithms for red and white cells in stool microscopic images.In the preprocessing part,this paper mainly carried out the smoothing and denoising,studied the Gauss filter and the anisotropic diffusion smoothing filter algorithms.The anisotropic diffusion smoothing filter can not only smooth the images and remove the noise,but also effectively retain the important information of the images.In the segmentation part,image segmentation based on threshold,edge detection and improved Chan-Vese model have been studied in this paper.Considering the characteristicws that stool microscopic images have complex background and weak boundary,this paper proposed a synergetic combination 8 direction Sobel edge enhancement with the tensor field as the region texture attribute in order to remedy the clear cell's fuzzy boundary and keep the cell's inherent texture.Compared with the traditional Chan-Vese segmentation model,the improved Chan-Vese segmentation model improves the segmentation precision obviously.In cell's features extraction and cell's classification part,shape features,statistical features and texture features are extracted and combined firstly,there are four classification algorithms are supported next,for example,SVM,decision tree,random decision forest and deep forest classification algorithms.In this paper,we used random decision forest to classify and identify red and white cells in the images,and improved classification effect by adjusting the number of training features,the combination of feature subset and the number of decision trees.At last,this paper compared and analysed the effects from the various image processing techniques and methods.Experiments show that the improved Chan-Vese segmentation model and the random decision forest classification method have higher resolution ability and stable optical environment adaptability for the classification and recognition of tangible objects instool images.A software framework which automatic identify the red and white cells in stool microscopic images was developed based on the relevant algorithms mentioned above.