Abstract: The shape classification and recognition of planar curve plays an important role in computer vision. In the shape analysis of stress surface contour of 3D footprint, the consistency of datum plane cannot be strictly controlled, so a descriptor insensitive to perspective transform is needed. In view of this, a one dimension polar distance sequence is obtained first from the closed planar curve and then the auto regressive model of this sequence is used to extract feature vector. Finally, the feature vectors are used to shape classification and recognition of planar curve. Compared with traditional algorithms, the features extracted using our approach have many advantages, such as invariance to sampling start point, robust to noise, less computation, and convenient to use in shape classification. Qualitative analysis and experimental results prove that the features extracted are insensitive to space rotation of a small angle. Experimental results also show that the new approach works properly well in confirming the identity of region contours of two footprints stress surface, and can be used in recognition and classification of 3D footprint.