Research And Realization On Feature Extraction Technology And ID3 Recognition Algorithm On Cell Image

The automatic detection and recognition of cell image is active in biomedical engineering domain, and makes great effect in theory and practice. Realization of the automatization can increase the efficiency of medical inspection and gets doctors rid of multifarious and repeated works, and it would enhance the authority of the inspection, no matter what inspector's knowledge and experience.Some works were done to cell image feature extraction and ID3 algorithm for decision-tree recognition:1. A new algorithm for image shape feature extraction by polygon approximation based on corner detection was presented. Four features, including direction grads variance, grads median, threshold area ratio and ration between numbers of corners and ports, were defined. These features can describe the different character of tube and crystal cells excellently.2. An improved method syncretising texture spectrum image inhancement and united Hu moment invariance description was presented. Simulation results showed that the features extracted could distinguish hypercythe and leucocyte by the method.3. A data standardization algorithm based on feature histogram equalization was carried out in uniforming various distribution to solve insufficiency of data separability caused by little knowledge to it. The method can quicken study speed of decision tree.4. ID3 algorithm was improved by using C-means clustering to avoid the trade of choosing the feature which has more choices. It made decision tree efficient and its application exact greatly.5. 25 features were extracted in all for describing the image shape and texture to build up feature bank, and the cell automatic recognition system was accomplished, which has been used in clinic fulfilling the request to recognition rapidity and precision.The experiments showed that the recognition rate of cell automatic recognition system increased by using these algorithms in it.