Machine Learning and Script Recognition

Machine simulation to recognize handwriting has opened new horizons to improve human-computer interface and perform repetitive task of reading by machines. Despite nearly four decades of research, offline unconstrained cursive script segmentation and recognition remains an open problem. Currently, accuracy of offline cursive script recognition schemes is low particularly for touched script in data entry forms and computationally expensive for real world applications. This book presents latest machine learning techniques for script recognition from pre-processing to post-processing. Character segmentation is a focus of this book as poor segmentation significantly effects accuracy. Several interesting experiments are carried out and results are compared with existing pre-processing and post-processing techniques reported in the state of art using benchmark database such as NIST, IAM and CEDAR. Finally, latest techniques are integrated into a model script recognition system. Remaining problems are also highlighted along with suggestions and recommendations.