Why should you use Hu/Zernike in DCT/DWT images to do comparison? Hu/Zernike work best when there is an outline or silhouette of an object and that is the exact reason why Hu works on uint8. However, if you want, you can convert the image in frequency domain (DCT/DWT) to uint8 by applying a threshold (output will be a binary image). But the DCT/DWT coefficients of an image block may probably not have a discriminating shape hence use of Hu may not work. Further Zernike moment is invariant to rotation, but if you are going to compare DCT image then the orientation will be a critical fact for the comparison (because same shape in different orientations in DCT domain implies totally different image content in spatial domain). Therefore Zernike moment is certainly not an option here. The actual answer depends on what you are going to do with that comparison.

Thanks for the reply, actually i want to find similar blocks in the image. Hu is working good in uncompressed image but dont work in compressed image. DCT an DWT may help in finding similar blocks even in compressed images, so trying to work on them.

Then discard some negligible coefficients (ex. coefficients which belong to extremely high and low frequencies) and apply PCA to further reduce the dimension of the set of remaining coefficients. Then use the PCA projection of remaining coefficients to do the comparison. You can even do the same even without removing the negligible coefficients because PCA itself will remove non-significant features once the principle components are discarded based on their eigen values.

Also you can try Histogram Of Gradients which may work in your case.