Two-photon fluorescence microscopy is a powerful technique to obtain the stacks of neuronal individual or population morphologies deep inside brain tissue in vivo. However, the stacks often suffer from increasing noises with depth because of light scattering of specimen and optical distortion of microscopic system. Therefore, deconvolution becomes a more useful and a crucial approach to restore the original details of neuronal structure in fluorescence images. Since Richardson-Lucy deconvolution algorithm is appropriate for Poisson process of microscopy but sensitive to noise, we propose a scheme that it pre-filters noise via Perona-Malik nonlinear anisotropic diffusion before performing regularized Richardson-Lucy deconvolution algorithm. In contrast to other restoration approaches, the preliminary denoising of Perona-Malik diffusion model provides a better trade-off between noise reduction and edge preservation, and helps to following regularized Richardson-Lucy deconvolution procedure. Experimental results have shown that proposed scheme is effective and robust for restoring noisy two-photon fluorescence images.