一种新的颜色相似度定义及其计算方法

A NEW DEFINITION OF COLOR SIMILARITY AND ITS COMPUTATIONAL METHOD

传统的基于直方图的颜色相似度计算方法的优点是与图像旋转、平移和尺寸变化无关，但是其缺陷也非常明显，例如两幅看起来颜色非常相似的图像，如果它们的颜色直方图交集为空或者直方图之间距离很大，那么相似度为零或者很小，这显然不符合人的感知．为克服此缺陷，作者首先提出了两幅彩色图像之间颜色相似度计算公式，然后给出基于颜色直方图的新的相似度计算公式，它既保留了用直方图计算相似度的优点，同时克服了其缺陷．实验结果表明。

Traditional methods to compute color similarity using color histogram own the following advantages: the similarity is generally invariant to translation and rotation of the images and scale if using the normalized histogram. But their shortcomings are obvious: if the intersection of the two histograms is null or the distance between the two histograms is large, then the similarity may be zero or very small no matter how the two images look similar in color. In this paper, a new way to compute color similarity between two color images is presented at first, then a simple formula of similarity computation is proposed based on 3D color histogram, which retains the advantage while removes the disadvantage of the traditional methods. Experiments show that this method can provide satisfactory retrieval results.