基于属性关系直方图统计的线状纹理图像检索方法

Linear Texture Image Retrieval Based on Attributed Relational Histogram

提出一种基于属性关系图ARG描述的线状纹理图像检索方法 .针对一般的ARG图匹配算法运算量大、检索速度慢的问题 ,在用ARG描述线状纹理特征的基础上 ,通过计算纹理基元属性关系直方图之间的归一化距离来衡量图像的相似度 ,大大提高了运算速度 .应用于鞋底花纹图像库的实验结果表明 ,该方法对于线状纹理特征具有较强的描述能力 ,对于平移、旋转和伸缩具有较好的不变性 ,检索速度和检索结果均能满足应用要求 .

Attributed Relational Graphs are used to describe linear texture characteristics. Three structural attributes, which are invariant to the change of scale, rotation and translation, are introduced to represent structural relations between each pair of texture primitives. To measure the similarity of two texture images, we represent multidimensional histogram of the structural attributes for each image and calculate the distance between two histograms. Compared to conventional matching algorithms, this algorithm can reduce calculation complexity largely. Experimental results from the application of sole pattern image database show that the proposed method can describe linear texture characteristics very well. The retrieval speed and performance are satisfying.