改进的二次型直方图距离度量

Improved quadratic-form metric for histogram distance measurement

为增强直方图距离度量的辨别力,对二次型的直方图距离度量问题进行了研究.首先,对传统的二次型距离及其存在的问题进行了简要评述;然后,通过对传统二次型距离度量函数进行改进和构建新的相关矩阵,提出了改进的二次型直方图距离度量准则;最后对所提距离度量的快速计算以及高维直方图的距离度量问题进行了讨论.改进的距离度量准则充分考虑了直方图子区间之间的相关性.通过形状匹配、图像检索以及视觉跟踪等一系列实验测试,验证了所提距离度量准则的有效性和优越性.

In order to improve the discrimination of histogram distance measurement, the problem of histogram distance measurement based on the quadratic-form metric is researched. Firstly, the traditional quadratic-form metric and its disadvantage are briefly reviewed. Then, an improved quadratic-form metric for histogram distance measurement is proposed by modifying the traditional quadratic-form distance function and the correlation matrix. Finally, the fast computation of the proposed metric and the measurement of high dimension histograms are discussed. The proposed metric takes a full consideration of the correlation between histogram bins, which increases its discrimination. Experiments based on shape matching, image retrieval and visual tracking validate the effectiveness of the proposed metric and its superiority over the traditional ones.