ContourDetect[m] finds zeros and zero crossings in a numerical matrix m, returning a sparse array.
ContourDetect effectively chops values smaller in magnitude than delta.
ContourDetect by default treats all eight pixels surrounding a given pixel as adjacent.
The option setting CornerNeighbors->False treats only the four pixels in the coordinate directions as adjacent.
Using the option setting CornerNeighbors->None, ContourDetect operates on the dual grid whose pixels correspond to the corners in the original image, thereby reducing the dimensions of the resulting image by one pixel.