一种新的形状描述与识别方法

A New Method of Shape Description and Recognition

在形状上下文基础上,提出一种新的形状描述符。该形状描述符以形状质心为参考坐标原点建立对数极坐标,并由质心与边界样本点之间的距离及其内角两维构成,其中内角采用质心与边界点连线与轮廓切线之间的切角。这种新的形状描述子计算简单,能较好地区分不同的形状,且具有平移、缩放和旋转不变性,适合使用神经网络来作为识别分类器。最后以神经网络算法作为分类器,在MNIST手写数字数据库和Kimia图像数据库进行了实验验证。实验结果显示:该方法在单目标形状图像检索中取得了较好的检索效果,同时显著地减少了检索所需的时间,适用于较大型图像数据库的检索任务。

Based on the shape context,a new shape descriptor was proposed.The shape descriptor was based on the shape of the center of mass to establish the polar coordinates of the reference point,and was formed by the distance and inner-angle between the centroid and boundary samples,and the inner-angle used the tangent-angle between the centroid and the boundary line and contour tangent angle.This new shape descriptor is simple to calculate,and can be good to distinguish between different shapes,and are invariant to translations and scaling,and is very suitable for using neural networks as recognition classifier.At last,the neural network algorithm was used as the classifier,and experimental results in MNIST handwritten numerals database and Kimia databases were shown that the method has achieved good results in single object shape image retrieval,and significantly reduces the search time,and is suitable for large image databases retrieval tasks.