Combining Analytical and Holistic Strategies for Handwriting Recognition

In this paper, a study is conducted on combining analytical and holistic strategies for handwriting recognition. Even though the big majority of the recent high recognition rate systems adopts analytical strategies, physiological scientists suggest that the holistic strategy is the key for realizing near-human performance. In what we believe is a fresh perspective on handwriting recognition, combining the two strategies results in improving recognition rate. The concept and the analysis of combining the two strategies is the first contribution of this work. The proposed approach is applied to an Arabic handwriting recognition system. We use a hidden Markov model (HMM)-based analytical recognizer, and four different methods combined as a holistic recognizer. Two of the used holistic recognition methods are novel: a method based on estimating the probability of writing in each pixel, and a method that uses the Hausdorff distance. The third holistic method is a connected components-based approach, and the fourth one uses dynamic time warping (DTW) with two modifications. These four methods and their use as a holistic recognizer are the second contribution of this work. The proposed technique is tested using the IFN/ENIT database. The combination between the holistic and the analytical recognizers has led to an evident recognition rate improvement ranging from 5% to 16% over the use of the analytical recognizer alone.