Wavelet coefficients prior statistical models of image have been studied widely in the Bayesian-based image processing scopes. In this paper,we derive a precise prior statistical model based on full Bayesian neural network (FBNN). The parameters of the model can be estimated empirically from a sample image set by modern particle samplers (Montel Carlo) methods. The simulated results based on the prior models of single scale and parent-children scale show the model makes it possible to exploit the dependency between the scales. Furthermore,a novel image denoising method based on scale prior particles sampled from the fitted the single scale and parent-children prior models produces the high quality visual effects and peak signal-to-noise ratio (PSNR).