Histogram-Based Image Retrieval Keyed by Normalized HSY Histograms and Its Experiments on a Pilot Dataset
Masaki Takeishi, Daisuke Oguro, Hisakazu Kikuchi, Jaeho Shin · 2018
As a proof of potential advantages of color-distribution similarity indexes, a toy system of content-based image retrieval (CBIR) system is developed and is operated on a pilot dataset. The CBIR system is keyed by a set of low-level features of normalized color histograms in HSY. The color distribution similarity is measured by Jensen difference between a pair of feature vectors. The dataset for experiments was automatically collected by our own web crawler from the world wide web (WWW). Image retrieval experiments were played by the proposed system and Google Image Search. A few comparisons of their results are given as a proof of effectiveness of the color distribution similarity that is capable of increasing the client's satisfaction with respect to color-oriented perceptual feeling.