ABSTRACT
In the previous chapter, we discussed automatic image annotation where a given image is labeled with text describing its contents. In restricted domains, image annotation can be a classification task if annotation is just a class label from a constrained set of classes. Classifiers trained with training set image features will be used for the prediction of unseen images. Many classification algorithms are available such as K-nearest neighbor (KNN), neural network, decision tree, Bayesian network, and support vector machine (SVM). It is hard to say which classification algorithm is better. We can only say one classification algorithm is better than
others for a specific problem. KNN is a very popular classification algorithm having good performance and short period of training time. Various KNN algorithms have been published. SVM is a new classification algorithm compared to other classification algorithms, and many research papers have shown that SVM can produce better classification results than other algorithms. In this chapter, we study different KNN algorithms and SVM in the medical image classification domain. Performance is analyzed based on classification accuracy. We present a data resampling method to solve the data imbalance problem. We also present a modified evidence theory-based KNN algorithm. Results show that resampling is helpful to improve the classification accuracy of KNN classifiers, and our modified evidence theory-based KNN algorithm outperforms other KNN algorithms discussed in this chapter. The 10,000 fully classified medical radiographs used for this study are from ImageCLEF (Cross Language Evaluation Forum) 2006. (See also [1-6]).
