Medical Image Annotation and Retrieval by Using Classification Techniques

Given the rapid increase in the number of medical images, the process of image retrieval is considered an effective solution that can be used in the automatic search and storage of images. Content-based image retrieval is considerably affected by image classification, also called image annotation. The performance of image annotation is significantly affected by two main issues, namely, automatic extraction for image features and the annotation algorithm. This study addresses these issues by constructing a feature vector from the extraction of multi-level features. Two machine learning techniques are used for evaluation. The K-nearest neighbor and support vector machine methods of learning machine are employed to classify images. Image CLEF med2005 is used as the database for the classification approaches. Furthermore, principal component analysis is utilized thrice to decrease the length of the feature vector. Results demonstrate that the accuracy is significantly improved compared with those of similar classification approaches related to the same database.