Recuperación de imagenes

Solo disponible en BuenasTareas
  • Páginas : 16 (3863 palabras )
  • Descarga(s) : 0
  • Publicado : 28 de noviembre de 2011
Leer documento completo
Vista previa del texto
2nd WSEAS Int. Conf on COMPUTER ENGINEERING and APPLICATIONS (CEA'08) Acapulco, Mexico, January 25-27, 2008

Content-Based Image Retrieval Using Wavelets
L. FLORES-PULIDO1, O. STAROSTENKO2, D. FLORES-QUÉCHOL1, J. I. RODRIGUES-FLORES1, INGRID KIRSCHNING2, J.A. CHÁVEZ-ARAGÓN1 1 Universidad Autónoma de Tlaxcala, Visual Tech. Lab. Apizaco, MEXICO, 2ICT Lab., Research Center CENTIA, Universidad de las Américas-Puebla, Cholula, Puebla, 72820, MEXICO,
Abstract: - This paper presents a novel approach for content-based image retrieval (CBIR) that provides the analysis of visual information using wavelet coefficients and similarity metrics. This approach has a better performance than well-knownRedNew CBIR system based on image indexing and retrieval using neural networks. A principal goal of this report is precise analysis of the four families of wavelets and three techniques for computing similarity. The best Symlet transform and similarity metrics based on Euclidian distance have been adopted in a proposed system called Image Retrieval by Wavelet Coefficients IRWC. In order to test aproposed system the recall and precision metrics used for evaluating the performance of CBIR facilities have been used on base of the standard COIL-100 image collections. The obtained results show the increment of retrieval efficiency up to 92% without additional increasing a processing time. Therefore a proposed approach may be considered as a good alternative for design of new CBIR systems.Key-Words: - Image processing, visual information retrieval, similarity metrics, wavelets

1 Introduction
In well-known content-based image retrieval systems the extraction of image features is the principal procedure used for image indexing and interpretation [1], [2]. There are a lot of reports about novel approaches and methods for searching, retrieval, indexing, and classification of visualinformation on base of analysis of low-level image features, such as a color, texture, shape, etc. [3]. These methods sometimes are slow and too complex for design of real-time applications. Additionally, CBIR systems use only low-level feature vectors, which do not provide mechanisms to represent a meaning of images. Another significant problem of image retrieval process is computing similaritybetween feature vectors of visual query and images, which are candidates to be retrieved. The comparison of the feature vectors without their adjustment and normalization frequently is not fast and convenient way to find the matching between them. That is why the feature vectors must be converted to other domains for simple and efficient image characteristics extraction, indexing, and classification[5]. Among the commercial CBIR systems that may be used as prototypes for development of novel retrieval techniques, CIRES (Content Based Image REtrieval System) is the one of efficient facilities that provides features retrieval, such as a structure, color, and texture of image combining them with
ISSN: 1790-2769

user specifications of importance of these features in a query [2]. SIMPLIcity(Semantics-sensitive Integrated Matching for Picture Libraries) provides image retrieval from the Web using texture, indexing by clustering of image segments, and feature vectors is generated by wavelets transform [6]. Another CBIR facility for classification of segmented images combining neural networks and wavelet matching is known as RedNew system [7]. This system provides a region growing usingmultiresolution in YIQ color domain applying Jacobs’ metrics for computing similarity between retrieved images and visual query. After testing this system some its disadvantages have been detected, such as a low percentage of relevant retrieved images and necessity of training the neural net for new classes of visual information. In this research we propose to apply a novel approach for RedNew...
tracking img