This article proposes a content-based image retrieval (CBIR) system that employs an informative pattern-based descriptor. Recent literature has reported the development of efficient local-pattern-based descriptors, including the local vector pattern (LVP). This article extends the LVP formulation by making it more computationally efficient and informative. In the extended LVP-based extraction process, the global index angles are determined using the mutual information between the patterns, which are obtained from a pair of indexed angles. Thus, the Proposed LVP (PLVP) no longer requires a step to identify patterns in every indexed angle found in the querying phase of the CBIR system. A CBIR system with the PLVP is developed in this article, and the system and its associated methods are tested using data from a benchmark texture database and a natural image database. A performance comparison of the PLVP and traditional patterns, such as the local binary pattern (LBP), completely modeled local binary pattern (CLBP) and local tetra pattern (LTrP), is conducted using the CBIR system. The experimental results reveal the superiority of the PLVP in terms of precision, recall, F-score and computational efficiency.

No abstract