Recovering Dynamic Information from Static Handwritten Images

Abstract

This paper proposes an efficient method for recovering dynamic information from offline single-stroke hand drawing images. This method makes use of both the local analysis and global smoothness calculation. At first, a graph model is built from the skeleton. Then, odd degree nodes are resolved in a probability framework to detect the double-traced/terminal segments, and even degree nodes are analyzed by the node traversing ride (NTR). We estimate the probability of two strokes being contiguous pair by PCA based angle calculation. Then, double-traced lines are identified. Finally, we calculate the smoothness for each of the possible paths by SLALOM approximation and select the smoothest one. Experiments show that our method works successfully on cursive hand drawing images.