Local Versus Global Features for Medical Image Retrieval

Abstract

-Medical images consists a formulaic composition of representation for each modality and anatomic region. Medical image categorization and retrieval is one of the most important aspects in medical field to provide similar looking images of an anatomic region for a better decision making in decease identification. Design of such search engine certainly improves the health care system. In this paper we experimented by extracting visual, textural and frequency attributes as the features for the query and database images and analysed the performance of retrieval process. The global features extracted from the whole image and the zone based local features computed from individual regions of the image. When the feature set certainly becomes large in dimensions then optimal features will be selected based on the dimensionality reduction algorithms, this work also contrasted the usage of principal components and singular value decomposed Eigen values for both global and local feature sets. The retrieval efficiency of the system is expressed in terms of mean average precision at 100% recall rates. IRMA 2008 dataset was used for this work. An empirical evaluation of the current implementation illustrates that local features significantly improve retrieval performance. The locally extracted features contain more information about the image objects and significantly worked better over global features.