Automatic Detection of Lumina in Mouse Liver Immunohistochemical Color Image Using Support Vector Machine and Cellular Neural Network

A novel method for automatic lumina detection is proposed according to the characteristics of immunohise chemical color images for mouse liver. Firstly, the classification model of support vector machine was generated by the trained sample pixels based on the texture features: entropy, standard deviation and range in G component of the original color image. Secondly, the trained model is used to classify all pixels in image. Thirdly, Extracting Closed Domain Cellular Neural Network and median filtering were applied in turn to eliminate the misclassified pixels in last step. The expert´s evaluation of our method showed that our method can get 76% classification accuracy, which are comparable to an pathologist´s judgement.