We consider image denoising via convolutional sparse coding with weighted \ell-1 penalization, and investigate the rationale behind the weighting scheme based on the reciprocal correlation between the dictionary and the image. We show that this weighting scheme, which has recently been proposed for convolutional sparse coding, yields, in case of orthonormal dictionaries, weights that are very close to the oracle weights in WaveShrink, i.e. the MSE-optimal soft thresholds. Furthermore, our empirical analysis shows that in the convolutional case, both weighting schemes achieve comparable denoising quality, providing a substantial improvement over the standard uniform weights.