Adding high noise to the image makes it high frequency everywhere, so the matrix won't become sparse after the quantization divide. Since it won't become sparse, it can't be re-represented in a more efficient structure. Thus, it can't be compressed as much.
Adding high noise to the image makes it high frequency everywhere, so the matrix won't become sparse after the quantization divide. Since it won't become sparse, it can't be re-represented in a more efficient structure. Thus, it can't be compressed as much.