Ultrasound image deconvolution has been extensively explored in the literature as a powerful tool to enhance the spatial resolution and the contrast of ultrasound (US) images, or to estimate the tissue reflectivity function from the radiofrequency (RF) data. Its principle generally uses the first order Born approximation, which leads a linear model relating the RF images to the tissue reflectivity function and the point spread function (PSF). Non linear interactions between the US waves and the tissues are the origin of harmonics in the received signals allowing to extract the so-called harmonic image. This paper deals with the joint restoration of fundamental and harmonic images by taking into consideration the axial variation of the PSF within a block-wise restoration process. The interest of the proposed method is shown through phantom and in vivo results and compared to joint restoration methods considering a spatially invariant PSF.