Boundary is an important characteristic to describe a shape in computer vision and image retrieval. A normalized Canonical Polygonal Representation (NCPR) which represents uniquely a polygonal boundary was introduced in the previous study. The objective of this study is to show 1) how a shape feature can be obtained from NCPR, 2) how to utilize this shape feature in shape comparison, 3) to explore effects of the feature parameters on shape comparison, and 4) to explore robustness and closeness of some similarity measures to visual similarity while using the feature proposed. Experimental results show that the approach proposed is robust and discriminative among similar and dissimilar shapes.