I assume that canny and sobel could also detect the document edges better if a different kernel to image resolution ratio is choosen. One way this could be don

I assume that canny and sobel could also detect the document edges better if a different kernel to image resolution ratio is choosen. One way this could be done is downsampling the image and then use sobel or canny edge detection.