Text detection using connected component labeling

Question

I'm creating a project where I have to detect text regions in an image and localize text lines. This I have not yet implemented, in this project after the initial image pre-processing I will generate an edge map and then use CCL (two-pass algorithm) to localize text lines.

My question is after the edge detection part all edges including those that are not text will be detected, so how do I filter out the non-text regions? Please, help this is the first time I had to make something like this.

Excuse my English.

Answer 1

I hope you are talking about Optical Character Recognition Algorithm...Here is a link to sample code in matlab OCR

You did not specified the platform you are working on, if it is opencv here is a link to simple tutorial OCR_opencv

Answer 2

You can use Gaussian filter to remove noises then Sobel filter to detect edges after that use non maximum suppression to eliminate weak edges and finaly use Hysteresis thresholding to make the edges white and background black:)