Research of Global Features of 3D Color Histogram with Similarity and Dissimilarity Metrics for Image Processing

Abstract

A review of the global feature color comprising of seven different metrics are discussed here. The color is the most powerful feature for describing the images. This study contains four sections. The first section explains the related techniques used in various papers. The second section explains the two different kinds of metrics. (1) Similarity metrics such as Cosine and Correlation and (2) Dissimilarity metrics such as Euclidean, Manhattan, Bhattacharyya, Chi-Squared and Intersection. The third section explains experiment results using CALTECHUCSD Birds-200 image library. The fourth section gives the conclusion and future work. In this experiment, the query image can be divided into trained (indexed) or untrained (non-indexed). In the similarity metric analysis, the experimental results show that the cosine similarity gives better similarity score than correlation. Similarly, in the dissimilarity metric analysis, the Bhattacharyya gives a better result than other distance metrics.