A novel approach for Kannada text extraction

Popularity of the digital cameras is increasing rapidly day by day because of advanced applications and availability of digital cameras. The detection and extraction of text regions in an image is a well known problem in the computer vision. Text in images contains useful semantic information which can be used to fully understand the images. Proposed method aims at detecting and extracting Kannada text from government organization signboard images acquired by digital camera. Segmentation is performed using edge detection method and heuristic features are used to remove the non text regions. Kannada text identification is performed using the structural feature boundary length of the object strokes. Rule based method is employed to validate the objects as Kannada text. The proposed method is effective, efficient and encouraging results are obtained. It has the precision rate of 84.21%, recall rate of 83.16% and Kannada text identification accuracy of 75.77%. Hence proposed method is robust with font size, small orientation and alignment of text.