Precise and Fast Form Identification Method by Using Adaptive Base Lines for Matching

Conventional form identification methods have been based on the normalization of an input image. So, if the base for normalization is different from that of the true model, it is difficult to identify its form. In this paper, we propose a form identification method, which prevents the difference from spreading throughout the process. In the method, the local ruled line structures are analyzed exhaustively by varying a pair of base lines of an input image and a model. The process is realized efficiently by generating the correspondence possibilities between ruled lines, and grouping these possibilities. We registered 100 models with a dictionary, and experimented on form identification under the various conditions. The result shows that the method has high accuracy and practical processing speed.