Hi,

you can obtain the gradient direction for each pixel by using the second output argument of the "imgradient"  function (in image processing toolbox):

[Gmag, Gdir] = imgradient(I)

The principle behind the function is to compute finite differences in x and y directions, to store the result in two arrays with same size as original image, and then to combine them by using atan2 function.

Thank you my colleague David, but I could not understand the last speech "to store the result in two arrays with same size as original image, and then to combine them by using atan2 function."

Also, these arguments, Gmag and Gdir in X-direction or Y-direction?

I don't have the program currently, but it is easy to implemet as listed below,

1. Convolution image with [-1 0 1] and [-1 0 1]' to get directional derivative in x and y direction as dx, dy.

2. then the direction can be calculated by atan2(dy,dx).

Gmag and Gdir are both computed from the directional gradients in X and Y directions. 

Chao Xing provided a good summary of the method.

Maybe the detailed code can help:

img = double(imread('rice.png')); % convert to double to avoid truncation problems with gradient on uint8 data

dx = imfilter(img, [-1 0 1], 'replicate'); % gradient in x direction

dy = imfilter(img, [-1 0 1]', 'replicate'); % gradient in y direction

Gmag = hypot(dx, dy); figure; imshow(Gmag, []);

Gdir = atan2(dy, dx);

hue = (Gdir + pi) / (2 * pi); % convert angle to [0 1] interval

hsv = cat(3, hue, ones(size(img)), min(Gmag/100, 1)); % compute color representation of directional gradient 

rgb = hsv2rgb(hsv); 

figure; imshow(rgb)

what is the meaning of angle of direction?