Performance Analysis of the Faugeras Color Space as a Component of ColorHistogram-Based Image Retrieval

摘要

The use of color histograms for image retrieval from databases has beenimplemented in many variations. Selecting the appropriate color space for similarity comparisons is an important part of a color histogram technique. This paper serves to introduce and evaluate the performance of a color space through the use of color histograms. Performance is evaluated by correlating the similarity results obtained from various color feature vector techniques (including color histgramming) to those gathered through a human perceptual test. The perceptual test required 36 human subjects to evaluate the similarity of 10 military aircraft images. The same 10 images were also compared via the color feature vector techniques. The results obtained for the Faugeras color space are compared against those of the Red, Green, Blue (RGB) and Hue, Saturation, Value (HSV) color spaces. While the correlation results for the Faugeras color space were unexpected and unfavorable, a Pearson correlation coefficient of 0.91 was obtained for the HSV space suggesting that HSV is an excellent color space for judging color image similarity. A discussion of the Faugeras space's performance and future research directions are presented at the conclusion of the paper.