Abstract: Shape retrieval has been a challenging issue in computer vision in the last decade. The measurement of the shape feature histogram distance is an important factor to evaluate the merits and demerits of the shape retrieval algorithm. This paper introduces pyramid matching, which is popular in image classification, into shape matching. Different from other traditional histogram measurement algorithms, pyramid matching divids the contour of the shape into bins, assigns weight to each bin and counts the characteristics of each bin. And then it calculates the weight of the feature, and measures the similarity of shapes by using contour features. The proposed algorithm has been tested on different shape databases, and the performance is superior to many other methods.

Key words: shape retrieval, contour feature, shape matching, pyramid matching, weight sum

摘要： 形状检索在计算机视觉中一直是一个具有挑战性的问题，其中对形状特征直方图距离的测量是评价形状检索算法优劣的一个重要因素。针对轮廓特征的直方图距离测量，算法引进一种在图像分类领域中应用广泛的金字塔匹配算法。不同于其他传统的直方图度量算法，金字塔匹配算法将形状的轮廓分成若干块，给每一块分配相应的权重，然后分别统计块中的特征，再计算特征的加权和进行相似度的测量。通过在不同形状数据集下实验，该方法能够有效地进行形状匹配和检索，且能得到较好的形状匹配精度。

关键词: 形状检索, 轮廓特征, 形状匹配, 金字塔匹配, 加权和