In this paper, a probabilistic model is proposed for high dynamic image’s tone reproduction. This novelmethod learns a distribution for local pixel energy of the tone. With the constraint of the gradient variation on theHDR image, an energy distribution is set up based on the similarity between the gradient variation on the HDR andthe LDR image. The probabilistic framework for the tone mapping operation is formulated into an energyminimization process by a Maximum A posteriori (MAP) deduction. It turns out that, the proposed methodgenerates LDR image with more visual information than the previous ones. Experimental results show that thisapproach is convincible and competitive, which can be applied in areas like advanced image editing, displayerdevelopment, etc.