The wavelet transform acts to segregate objects in function of their size. We apply this method on images of galaxies to decompose them into coefficients representing only objects of the same size. The total fluxes of the wavelet coefficients describe the cumulative power spectrum of spatial frequencies. Based on this spectrum, we propose a new parameter to quantify the galaxy texture. As expected, it remains small and quite invariant for early-type galaxies, while it covers a large range and takes larger values for late-type galaxies. Combined with a second parameter, our determination of the texture is able to successfully separate galaxy types. By thresholding the wavelet coefficients, we detect luminous lumps. In irregular galaxies, their radial distribution seems to show a double peak. This could be the trace of a privileged radial distance of strong star formation regions.