Efecto de la Transformada Motif en Desarrollo de Descriptores de Textura para la Extracción de Imágenes

In this paper, the effect of the Motif Transform to design and develop texture-based descriptors for the Content Based Image Retrieval (CBIR) task is analyzed. The Motif transform assigns a number of limited range to local region of 2×2 or 3×3 pixels, to summarize texture feature and reduce the dimension of the descriptor. To analyze and evaluate the retrieval performance of the Motif-transform based descriptors, the performance of four descriptors based on the Motif Transform and four commonly used texture descriptors in the literature are compared. The performance of each descriptor is assessed by means of the Recall-Precision curve and the Average Normalized Modified Retrieval Rank (ANMRR) using four texture databases. The experimental results show that the descriptors based on Motif transform provide acceptable performance, keeping sufficiently small dimension, which is an indispensable requirement for any CBIR system.