Wavelets families and similarity metrics analysis in VIR system design
Leticia Flores-Pulido, Oleg Starostenko, R. Contreras-Gómez, L. Alvarez-Ochoa · 2008
Abstract:- This paper presents an analysis of some novel approaches for visual information retrieval (VIR) system design which are based on extraction of wavelet coefficients and applying specific similarity metrics. Four families of wavelets and three techniques for computing similarity between queried and retrieved images have been tested using designed Image Retrieval by Neural Network and Wavelet Coefficients (RetNew) system. The best Symlet transform and similarity metrics based on Euclidian distance have been adopted in a proposed VIR system called Image Retrieval by Wavelet Coefficients (IRWC). Additionally, in order to evaluate a proposed approach and a novel designed system the recall and precision metrics used for analysis of the performance of VIR facilities have been applied on base of the standard COIL-100 image collection. The obtained results show the increment of retrieval efficiency up to 93 % without additional increasing a processing time. Therefore a proposed approach may be considered as a good alternative for designing new VIR systems. The obtained results allow facilitating the development of new methods for solving this still open problem of efficient image retrieval.