Shape classification using line segment statistics

• In this study, we use straight-line segment statistics for shape analysis.
• For each contour point, we consider a continuous portion with length equal to a pre-defined percentage of the contour size.
• Then, we compute the length of the straight-line segment between its extreme points.
• For the set of straight-line segments, we compute statistical moments (average and stardard deviation).

Contour shape description is an important field in computer vision. This is due to the fact that shape is an important low level image feature. In light of this, many approaches have been proposed in order to analyze it. Therefore, this paper introduces a very simple, yet efficient, shape descriptor based on straight-line segment statistics. For each contour point, we consider a continuous portion of the contour with length equal to a pre-defined percentage of the contour size. Then, we compute the length of the straight-line segment between its extreme points. For the set of straight-line segments, we compute statistical moments (average and standard deviation). Lastly, we perform this calculation for different lengths of contour portions. The proposed shape descriptor is a powerful tool for shape discrimination: it is robust (it can characterize a huge set of different classes of shapes) and is tolerant to variations in the shapes’ scale and orientation. Classification results of the proposed method overcome traditional methods found in literature, proving that it is an efficient tool for shape analysis.