Wavelets vs Shape-Based Approaches for Image Indexing and Retrieval

This paper presents a comparative analysis of some novel approaches proposed by authors for content based image retrieval (CBIR). One of them uses Two-Segments Turning Functions (2STF) and provides searching and retrieval of the multimedia documents within digital collections. Another technique retrieves images computing similarity between wavelet coefficients of querying and preprocessed images. For this purpose the Symlet transform has been used in designed system called Image Retrieval by Neural Network and Wavelet Coefficients RedNeW. However both of approaches operate with low-level characteristics processing color regions, shapes, texture, and they do not provide the analysis of image semantics. In order to improve these systems a novel approach is proposed that combines non-sensitive to spatial variations shape analysis of objects in image with their indexing by textual descriptions as part of semantic Web techniques. In the proposed approach the user's textual queries are converted to image features, which are used for images searching, indexing, interpretation, and retrieval. A decision about similarity between retrieved images and user's query is taken computing the shape convergence and matching to ontological annotations (if objects in image providing in this way definition of the machine-understandable semantics. In order to evaluate the proposed approach the Image Retrieval by Ontological Description of Shapes IRONS system has been designed and tested using some standard domains of images.