Provided you have the Image Processing Toolbox, code for Gaussian filtering of an image looks like this:
sigma = 2; % set sigma to the value you need
sz = 2*ceil(2.6 * sigma) + 1; % See note below
mask = fspecial('gauss', sz, sigma);
newImage = conv2(Image, mask, 'same');
The choice of the size of the mask (or kernel) array is a trade-off between truncation errors and computation time. The formula above is a reasonable compromise which also makes the mask size odd, so that the mask has a definite centre.
The 'same' option in conv2 makes newImage the same size as Image, which is convenient, but it assumes that Image is surrounded by zeros, so pixels near the boundary of newImage will be pulled towards zero.
If you are doing a lot of this, you may find that convolve2 (in the file exchange) saves time, as it is faster than conv2 for masks above a certain size. convolve2 also offers a 'reflect' option which produces output which often displays better than conv2(..., 'same') by making a symmetry assumption at the boundaries. filter2 should also be fast, but does not have the 'reflect' boundary option.
