Scale Invariant Descriptor for Content Based Image Retrieval in Biomedical Applications

Abstract

Content based image retrieval (CBIR) is an application of computer vision which tackles the problem of recovering images in large datasets based on a similarity criterion. The role of CBIR in biomedical field is potentially very important, since every day large volumes of different types of images are produced. An effective and reliable CBIR system can help the decision-making process and support the diagnosis made by the clinician, thanks to the possibility to analyze images similar to the one under test. Many successful CBIR systems use features based on local descriptors for image retrieval. In this work, a Bag-of-Words encoding paradigm based on the Scale Invariant Descriptor (SID) is used to extract robust features from the images. For the evaluation of the proposed technique, three datasets in biomedical field have been used: OASIS, which is an MRI dataset, Emphysema and NEMA, which are instead CT datasets. In order to evaluate the effectiveness and the reliability of the proposed technique also in other application fields, some experiments have been carried out on the ORL facial image dataset, used for biometric applications. The results show that the proposed technique outperforms or is comparable to state-of-art CBIR techniques.