The Research Of Stool Microscopy Image Recognition Based On Fuzzy Clustering

Automatic recognition of medical microscopy image analysis is focus and hotspot in biomedical engineering study. Decal detect is one of three usual medical practices, which can accurately determine the absence of inflammatory products, blood cells, parasite eggs and so on and are significant for the clinical diagnosis. Achieving clinical examination of automatic identification and cell image analysis can effectively improve the efficiency of the medical examination, and accurately provide scientific auxiliary information of decision for medical diagnostic. This thesis is centered on key technologies of stool microscopy image identification to deeply study involving five aspects such as image preprocessing, image segmentation, feature extraction, pattern recognition and fuzzy clustering. Algorithms have been experimented and analyzed to achieve the intelligent identification and classification of stool microscopy images.1. On image preprocessing, according to stool microscopy image’s features with numerous objects, complex background and noise interference to make analysis of the main factors of degradation and noise model. It combines with the characteristics of image in retaining the target image edge information so as to achieve a smooth image when filtering image background noise and impulse noise.2. On the image segmentation, firstly, it deals with image binarization, and using edge detection and threshold segmentation method to achieve a tangible component separation cell image segmentation based on features of the multi-information of microscopic cell images, multi-levels. It applies morphological algorithms based on edge connector to overcome cell edge fracture and adhesion phenomena. 3. On feature extraction, after the image is divided up, it makes reference to the morphological characteristics (area, perimeter, circularity, etc.) texture characteristics (contrast, variance, two order angular distance, etc.) to totally extract18kinds of characteristic parameters for laying the foundation of cluster analysis in order to determine the characteristic parameters of the physical components of the microscope.4. On pattern recognition, this thesis introduces the basic framework and assignment of visible part model identification of microscopy image. The final recognition algorithm is determined through comparing the performances of the BP neural network algorithm and decision tree algorithm5. On fuzzy clustering, the thesis puts feature parameter data obtained from extracting features in cluster analysis, and finally use ISODATA clustering algorithm to cluster parameters feature datasets. All classes after clustering are ones which are to be identified (WBC category, RBC etc). Lastly, the target component is determined by neural network pattern recognition methods and decision tree pattern recognition methods. This method can achieve the goal of classified identification and get better effects with no necessary identification of each individual that is to be identified.