Motif correlogram for texture image retrieval

Abstract

In this paper we present a novel, compact and effective method for extracting texture information from an image. We denominate this method as Motif Correlogram (MC), which computes the correlation between motif pairs of the same type. The proposed method was evaluated using different metrics commonly used in image retrieval, such as ARP (Average Retrieval Precision), ARR (Average Retrieval Rate) and ANMRR (Average Normalized Retrieval Rank). Also, the proposed scheme was compared with other texture descriptors, such as Steerable FIltres, Edge Histogram Descriptor (EHD) and two Co-occurrence Matrix-based algorithms: Motif Co-Occurrence Matrix (MCM) and Directional Local Motif XoR Patterns (DLMXoRP). The performance of the proposed method was evaluated using the Kylberg Dataset. The evaluation results show the proposed texture descriptor improves the texture image retrieval performance.