I think it's more typical to low-pass (i.e. blur) the image and then use a line-detection algorithm like the Hough transform. Properly deskewed text should have prominent horizontal white lines.