Advanced search
Start date

Integration of multiple metric spaces for similarity queries: applications in medical images


Nowadays, Database Management Systems (DBMS) must be able to manage complex data, such as image, audio and genetic sequences. For Content-based Retrieval on large collections of complex data, the similarity among elements is the most relevant concept, and it can be adequately expressed when data are represented in metric spaces or multidimensional spaces.However, the similarity is not measured directly on the complex data, but on features extractedfrom them representing the intrinsic content of complex data. Various features vectors canbe extracted for the same object, describing dierent aspects of the data. With images, forexample, features that describe shape, texture and color distribution of images are widely used.In the same way, can be used temporal information associated to images, or specic features ofan image domain. It is common that queries using combinations of feature vectors have betterresults than using only one. In this work, we explore the combination of multiple metric spaces,referring to dierent feature vectors, focusing specially on medical images datasets. We intendto develop methods for scaling multiple feature vectors representing the same object, usingintrinsic and semantic information associated to images. Furthermore, we intend to improvethe results of content-based retrieval, not just increasing response accuracy, but also seekingto identify and meet the user needs. We intend to include a determined degree of variety inquery results, without degrading the time of their execution. We intend to use multiple metricspaces, which can assume dierent roles during the queries. For example, we can use dierentfeature vectors in the computation of similarity and diversity. (AU)