Text Detection on Natural Images Using Mnemonic Cellular Automata

Textual information that resides in natural images is an important knowledge for indexing and retrieval purposes. In this paper, a new approach is proposed using Mnemonic Cellular Automata (m-CA) which strives towards detecting scene text on natural images. Initially, an edge map is calculated and consequently binarized. Then, taking advantage of the Hybrid Cellular Automata (CA) flexibility, the transition rules are changed and are applied in different consecutive steps. Initially, its rules partially depend on Coordinating Logic Filters (CLF) and the majority state. Moreover, in the final steps of the m-CA evolution the update rules are modified as the history of past evolution steps is incorporated into each cell. Experimental work on the ICDAR 2011 Robust Reading Competition dataset shows improved performance.