Intelligent Content Based X-Ray Image Retrieval using Speeded up Robust Feature Descriptors

In this paper, an intelligent CBMIR system developed to classify and search for relevant X-ray images is presented. Here, a total of 1750 X-ray images from Image Retrieval in Medical Applications (IRMA) database belonging to 25 different categories are used. These images are preprocessed and features are extracted using principal component analysis (PCA), wavelet transform and speeded up robust feature (SURF) descriptors. A bag of visual words or codewords are generated, from the features, to represent images in the database. This helps in forming the feature vector of the query image. Classification performance has been compared using three classifiers: support vector machine (SVM), k-nearest neighbor (KNN) and relevance vector machine (RVM). A 5-fold cross validation approach resulted in a maximum accuracy of 92.256% using SURF descriptors and a kernel based SVM classifier.