Abstract: To improve the efficiency of natural artificial object recognition and defect diagnosis, a method based on improved shape context which achieves the collected and the sample image matching is proposed. The collected image’s contour is uniformly sampled. The sampled points are processed by the improved shape context. By using the Hungarian match between the collected and the sample image all the matched points are got. All the distances are computed between the matched point. The collected image is processed by TPS transformation, and the distances between the matched point are computed again. Based on the changing scale of the distance between matched points before TPS and after TPS transformation, this paper determines whether the sample image and the captured image are the images of same category. Based on the same category, the shape context distance is computed. By the matching cost of shape context, it measures how many points are matched to diagnose the image defect. This method achieves the image recognition and defect diagnosis together. The results are satisfying and will be implemented in the Helicopter Inspection System of high voltage power lines.

Key words: natural artificial object recognition, artificial object defect diagnosis, uniform sample, shape context, Hungarian match, Thin Plate Spline（TPS）

摘要： 为了能有效提高识别自然人造对象并诊断其缺陷的效率，提出了一种基于改进的形状上下文实现采集图像和样本图像匹配的方法。该方法对采样图像进行轮廓均匀采样并改进其形状上下文，将采样图像与样本图像进行匈牙利匹配，计算出所有匹配点间欧式距离。对采样图像进行TPS变换，计算匹配点间的欧式距离；根据TPS变换前后匹配点间距离变化比例判定样本图像与采样图像是否是同类图像；在同类图像的基础上计算形状上下文距离，根据匹配代价衡量匹配点集中匹配上的点数实现图像的缺陷诊断。该方法同时实现了图像识别与缺陷诊断，实验结果较好，能应用到架空输电线路直升机巡检系统中。

关键词: 自然人造对象识别, 人造对象缺陷诊断, 均匀采样, 形状上下文, 匈牙利匹配, 薄板样条（TPS）变换