Uso de saliências do contorno via esqueletização para recuperação de imagens

Abstract

Image retrieval is an offshoot of information retrieval, which has received growing attention. Within this new offshoot the area of shape-based retrieval has shown itself to be one of the most difficult tasks to perform successfully. It is therefore the purpose of this work to explore the main aspects of shape-based retrieval: the representation and the description of shapes, which aim at extracting the main shape forms, as well as the similarity measurements which are used in the comparison between two characteristic vectors which are the result of characteristic extraction. Various descriptors of different classes are covered giving a general view of how shape based retrieval can be carried out. When given further consideration these descriptors are able to point out some desirable aspects, with the invariance to geometric transformations being one of the principal ones. Therefore the descriptors are analyzed to their degree of invariance to transformations, also normalizations in the sense of making the descriptors invariant are analyzed. One new descriptor, contour saliences, in particular shows promising results as it is invariant to such transformations. For this reason the subject is widely discussed in this work, its retrieval is carried out through the use of skeletons being that each saliency is associated to an adjoining extremity, and this is obtained through the use of the image forest transform. A CBIR based on shape that uses salience is implemented and the final results are discussed.