It is totally fine, it is simply a sparse representation rather than a dense one. That is, it stores a count of each feature in each sample, regardless of whether or not it was found in that sample.
This thread has some great discussion regarding the evaluation of denoising results, unfortunately there is not really a "right" answer, but you should consider the questions raised there and be able to justify the results you use in light of these considerations 
sorry if that does not feel like a very helpful answer, I know there is nothing quite so unhelpful as "it depends" but in this case it really does. But your results do seem at least reasonable 