One way in psychovisual experiment to understand human visual system is to analyze separately contents of different spatial frequency bands. To prepare images for this purpose, we proceed to a decomposition of the original image by a wavelet transform centered on selected scales. The wavelets used are Difference Of Gaussians (DOG) according to works modeling the human visual system. Before rebuilding the visual stimulus, various transformations can be performed on different scales to measure the efficiency of the observer, for a given task, according to the spatial frequencies used. The problem is that if we use an incomplete wavelet basis during decomposition, there is a significant loss of information between the original image and the reconstructed image. The work presented here offers a way to solve this problem by using coefficients appropriate for each scale during the decomposition step.