基于感知特性的动态相似度量

Improving Perceptual Matching in Color Image Retrieval

在基于内容的图像检索中,常用的M inkow sk i距离还不能实现与视觉感知距离的精确匹配。依据韦伯-费克纳法则,提出一种基于感知特性的相似度量算法。通过构造动态感知因子,新算法能够有效地降低相关图像相关特征之间的距离,从而提高相似度量与感知系统相似评价的一致性。实验证明:新算法能有效降低相关图像在检索结果中相似排序的序数,有效地提高检索效率。

All researchers in CBIR （Content-Based Image Retrieval） domain try hard to match computer calculated distance with the distance sensed by human visual system. But, in our opinion, the present status is still far from achieving satisfactory matching. We aim to present a method that we hope can make some progress towards satisfactory matching. In the full paper, our method is explained in much detail; here we give only a briefing. We discuss the following two topics： （1） the shortcoming of Minkowski distance; （2） reducing the inadequacy of Minkowski distance with a dynamic perception-based distance based on Weber-Fechner Rule. Finally we present experimental results for two feature （color histogram） databases from a relatively large image database. We retrieved color images by our method and also by two Minkowski methods： L1 and L2. Here we give only the comparison results of our method with L1 method （better than L2）： for one feature database, the retrieval efficiency of our new method is better than that of L1 by 6.6% to 24.8%, the average being 16.7% ; for the other feature database, the retrieval efficiency of our method is better than that of L1 by 6.2% to 22.7%, the average being 9.7%.