Local Features and Kernels for Classification of Texture and Object Categories: A Comprehensive Study

摘要: Recently, methods based on local image features have shown promise for texture and object recognition tasks. This paper presents a large-scale evaluation of an approach that represents images as distributions (signatures or histograms) extracted from sparse set keypoint locations learns Support Vector Machine classifier with kernels two effective measures comparing distributions, the Earth Mover’s Distance ÷2 distance. We first evaluate performance our different detectors descriptors, well classifiers. then conduct comparative several state-of-the-art 4 5 databases. On most these databases, implementation exceeds best reported results achieves comparable rest. Finally, we investigate influence background correlations performance.