Part-based method on handwritten texts

This paper reports a trial of handwritten text recognition by a part-based method. The part-based method recognizes individual characters by their parts without considering their whole shape. This realizes great robustness to severe deformations. This robustness is also effective for text recognition. Especially, for handwritten texts whose segmentation into individual characters is very difficult by deep touching and heavy slant, the part-based method still can recognize them because it does not request segmentation results to provide their whole shapes. Experimental results using digit sequences proved this robustness.