Using skewness and kurtosis for iterating accurate thresholds ϖ_λj in the cleaning process (Sect. 2.3). The upper panels show histograms for the full image ℐ_λD at 350 μm (left) and for the decomposed images ℐ_λDj at the 9′′ and 36′′ scales (middle, right) before cleaning. The lower panels display histograms for the reconstructed residuals ℐ_λD   R (left) and for the residuals ℐ_λDj   R of the two decomposed images of the upper panels (middle, right). The vertical lines in the left panels indicate the standard deviation σ_λ (short dash, green) and 6   σ_λ (long dash, magenta) computed in the full image ℐ_λD. In the other panels they indicate the converged values of the standard deviation σ_λj and 6   σ_λj in the single-scale images ℐ_λDj, as well as the final thresholds ϖ_λj (solid, red). The histogram of the residuals ℐ_λD   R of the cleaning process, reconstructed from all spatial scales (lower left) has much greater symmetry and resembles a Gaussian distribution, whereas the histogram of the full image ℐ_λD (red, copied from the upper-left panel) is highly asymmetric. Both and have a value of 3.17 and the corresponding variable factors n_λj have the values of 4.52 and 4.09 for the two single-scale images. The width of the intensity bins is 1 MJy/sr in the left panels and 0.004 MJy/sr in all other panels.