Multi-level classifier design for tumor micro-image based on multi-feature fusion

In this paper, we propose a computer aided recognition system. With a series of approaches of image preprocessing and segmentation, extract the whole image feature namely global features, such as the number of cell also extract the single cell feature namely local features, for example cell area and so on. Then separately do the multi-feature fusion for global feature and local feature. We construct two level of classifier, the first level is global classifier based on the most-short distance which is designed according to the global features; the second level is local classifier based on decision tree which is designed according to the local features design. Firstly according to global classifier to judge whether the stomach epidermis is normal or not, if the image is recognized to be abnormal, the classification ends. Otherwise, we recognize the image by the local classifier once again. After two level of recognitions, The experiment result indicates that this classifier can enormously enhance the classification accuracy rate.