In order to achieve a better denoising effect, we analyze the advantages and disadvantages of the existing denoising model based on the partial differential equations in spatial domain, and propose a double-fidelity total variation denoising model on the basis of L p adaptive denoising model. We introduce the adaptive parameter as the weight of the fidelity item to control the fidelity intensities of different regions and improve the adaptive denoising ability. To overcome the 'ladder' effect in smooth areas, an adaptive gradient-fidelity item is added so that the denoising images have the similar gradient structure to the original images. The experimental results show that the proposed model has better effects in the peak signal to noise ratio, structural similarity, and visual effects.