We propose a quadtree segmentation based denoising algo- rithm, which attempts to capture the underlying geometrical structure hidden in real images corrupted by random noise. The algorithm is based on the quadtree coding scheme pro- posed in our earlier work [12, 13] and on the key insight that the lossy compression of a noisy signal can provide the fil- tered/denoised signal. The key idea is to treat the denoising problem as the compression problem at low rates. The in- tuition is that, at low rates, the coding scheme captures the smooth features only, which basically belong to the origi- nal signal. We present simulation results for the proposed scheme and compare these results with the performance of wavelet based schemes. Our simulations show that the pro- posed denoising scheme is competitive with wavelet based schemes and achieves improved visual quality due to better representation for edges.