目标识别的极指数栅格方法

在极对数坐标下常常通过傅立叶变换得到目标的轮廓不变量来识别目标，提取的特征量多且花费的时间长，因此提出了一种新的目标识别的极指数栅格方法，该方法首先将直角坐标中的目标映射到极对数坐标下，把包围变换中心的目标轮廓变换成一维目标曲线，然后提取曲线的结构特征，包括目标的跨度、目标曲线面积比率和目标曲线分布状况，这些结构特征具有旋转、缩放、平移不变性，用BP网络对3个二维目标进行学习和识别，实验证明，利用结构特征进行识别得到了较好的识别效果，并且花费的时间少，但是本方法仅适用于没有滚动和扭动的单个运动目标识别。

Object recognition using polar-exponential grid technique

Object recognition with contour invariant extracted by FFT is a regular method in polar_log coordinates, but the features extracted are more and the recognition time is lengthy, so a new polar_exponential grid technique was proposed for object recognition. First, an object in the Cartesian coordinates was mapped into the polar_log coordinates, and the closed contour of the object was transformed into a 1_dimension curve. Second, the structure feature of the curve was extracted, which included span of the object, area ratio, and distribution status of the curve. This rotation, scale, and translation invariant can be used in object recognition. Three 2D objects were learned and recognized by the BP network in the experiment. The recognition result using the structure invariant and the time it took improved over the contour invariant; however, it only can be used for a single moving object without rolling and twisting.