I assume that in the segmented regions has different label values and that the background region has value of 0.

In Matlab you can do the following:

saliencyForRegion1 = mean(saliency(segmented==1));

if you want this for all the regions that are different than 0:

for label=1:1:length(unique(a))-1

 meanSaliencyOfLabel(label,1) = mean(saliency(segmented==label));

end

Thank You Amitay Nachmani .How can i see the output image?i mean imshow gives no output

May be i miss understood the question but you asked for an average value per region that means that you get a scalar value for each region and thus it isn't an image.

i think my question is not clear.as seen in the image, i want to show the deer and calculate the mean of region (deer only). there may be 2-3 regions in the image.So i want to show each region separately and calculate the mean of each region.

To get an image of the saliency only for a specific region just do:

saliencyRegionLabel = saliency.*(segmented==label);

The avarge value of the saliency is:

 saliencyAvrageForRegionLabel= mean(saliency(segmented==label));

if you look up the graph based saliency toolbox in matlab on the web, you can use that to segment your region in a way that its pretty easy to just run a boxcar saliency indicator over. an example would be statistical noise via a local STDev estimate. I can point you to some change detection publications that explain the process.

rereading your other posts - why not just do a binary map of the region of interest with some morphological tool, then run a window of saliency estimates of your choice (see my prior comment) along the whole mxn matrix?