距离分布直方图及其在商标图案检索中的应用

Distance Distribution Histogram and its Application in Trademark Image Retrieval

形状是描述图象的重要视觉特征 ,它可以通过像素点分布在空间不同的区域而表现出来 .针对二值图象提出了一种基于区域的形状特征 ,即距离分布直方图 .它的基本思想就是通过统计图象中像素点在距离区域的分布情况来获得形状特征 ,其中 ,基准点的选择和距离区域的划分是两个重要的部分 .实验结果表明 ,距离分布直方图能够有效地刻画出二值图象的形状特征 ,并且具有非常好的平移、尺度和旋转不变性 .将其应用于商标图案检索 ,其检索结果符合人眼的视觉感受 .

Shape feature is a very important feature used to depict image, it is difficult to describe shape. This paper proposes 'distance distribution histogram', a region-based shape feature which uses the dimensional distribution information of pixels in binary image to describe the image shape. The basic idea is to get shape feature by calculating the distribution of pixels in different distance regions. In the method, the selection of datum mark and the partition of distance region are two important parts to assure that the feature can have good invariant performance under some rigid transformation. After the distance distribution histograms have been calculated for different images, Quadratic-form distance is then used to compare the similarity of the images. The experiments have been done in a database consisting of 3000 binary trademark images, the results show that this shape feature can depict the image well, and it has very excellent invariant performance under translation, scale and rotation. When the feature is used for binary trademark image retrieval, it is obvious that the retrieval results fit with human's visual perception well. From the experiments, it can be concluded that the distance distribution histogram feature is really an effective shape feature for binary trademark image retrieval.