A Combined Edge and Connected Component Based Approach for Kannada Text Detection in Images

This paper presents an approach for Kannada text detection from scene images. The sturdiness of our system comes from three facets. Initially, we used a morphological edge detection method. In this a threshold level is determined resulting in an edge image. Secondly a text feature filtering method is applied which removes maximum non-letter components. Also, the method results in majority of letter components that are within the range of mean height of the residual bounding boxes. Finally a text region binarization method is applied using a threshold from the residual bounding boxes. The performance of the proposed method was tested on our own collection of 200 kannada images and the results are satisfactory.