Non-Linear Segmentation of Touched Roman Characters Based on Genetic Algorithm

The segmentation accuracy of Roman cursive characters, especially touched characters, is essential for the high performance of Optical Character Recognition Systems. This paper presents a new approach for non-linear segmentation of multiple touched Roman cursive characters based on genetic algorithm. Initially, a possible segmentation zone is detected and then best segmentation path is evolved by genetic algorithm. The initial population is composed of each point column in possible segmentation zone. The individual coding, fitness function, crossover operator and mutation operator are also defined for this task. Experimental results on a test set extracted on the IAM benchmark database exhibit high segmentation accuracy up to 89.76%. Proposed approach can handle some complex types of touched cursive characters without special heuristic rules and recognition.