http://stackoverflow.com/questions/8438674/dense-sift-vs-sift

SIFT consists of both detection and description while dense sift only uses the descriptor in densely sampled locations. 

SIFT identifies interest points using Difference of Gaussian Filtering (DoG) before using Histogram of Oriented Gradients (HOG) to describe these interest points, however Dense-SIFT does not identify interest points, it simply divides the image into overlapping cells before using HOG to describe them. since they both use HOG they both produce 128 dimensional feature vectors.

Thank you for your answer Mr. Abasse Olaode.

So we can say that  dense SIFT is not scale invariant ??

Yes, I think dense SIFT is not scale invariant.