Shape recognition using skeletal critical points

In recent years, with the advances in imaging technologies and widespread use of the internet, a great number of images and video are present in the digital world. Because the number of images and video is very large, the need for searching, indexing and grouping operations is high. With the help of computer vision techniques, the images can be interpreted and their contents can be recognized for the mentioned operations. In this work, we presented a novel shape recognition framework based on medial axis graph. After extracting medial axis graph of a shape, we constructed a multi-dimensional distribution by calculating shortest path distances between each skeleton point and all of the points in the graph. Then similarity rates between these distributions is found by using a transportationbased distance function. We compared our results with results of similar works conducted in the past. Finally, the performance of the proposed method is increased by using feed forward neural network algorithm which provides a fair amount of improvement on classification results.