Character recognition of Kannada text in low resolution display board images using zone wise statistical features

Automated systems for understanding text in low resolution natural scene images of display boards are facilitating several applications such as blind assistants, traffic guidance systems, tour guide systems, location aware systems and many more. The text recognition at character level is one the important processing steps for development of such systems. In this work, a novel method for recognition of Kannada basic characters using zone wise statistical features is proposed. The method works in two phases; In the first phase, the zone wise statistical features are obtained from training samples and knowledge base is constructed. During testing, the test image is processed to obtain zone wise statistical features and character is recognized using nearest neighbor classifier. The method has been evaluated for 1043 samples and achieves an average recognition accuracy of 83.49%. The method is robust and insensitive to noise, blur, variations in font size and style, uneven thickness and varying lightning conditions.