An assessment of MPEG-7 visual descriptors for images of maize plagues and diseases

Abstract

Feature description is a fundamental process in the analysis of images for content-based retrieval and classification, among other tasks in which the image feature descriptor should have enough discriminative power to differentiate similar from dissimilar images according with a distance measure. Although several descriptors have been proposed for a variety of images, the challenge is their suitability to solve these tasks efficiently. One proposal is the set of standard MPEG-7 visual descriptors. We address their suitability to efficiently describe plagues and diseases in images of maize plants. The importance of this crop is its worldwide relevance for human and animal consumption. Experiments for similarity search queries using a set of distance measures, show that the Color Structure Descriptor with the Bray-Curtis distance is the most efficient and provides around 68% precision in most cases.