Local color oppugnant quantized extrema patterns for image retrieval

Abstract

A new image feature vector named, local color oppugnant quantized extrema pattern (LCOQEP) is proffered in this paper. In the proposed method, color-texture information between two oppugnant colors planes, RGB and HSV is extracted. The proposed approach is different from the local oppugnant color space extrema pattern that explores the directional extrema in RV, GV and BV spaces of an image. In LCOQEP approach, quantized extremas from the oppugnant planes are extracted. Performance evaluation parameters such as precision, recall, average retrieval precision and average retrieval rate are employed to ascertain the efficacy of the devised method. Standard datasets such as Corel (1k, 5k, 10k) and ImageNet-25k are used for experimentation. A substantial improvement in the retrieval performance is observed.