Three fractional-order TV-L2 models for image denoising