The invention discloses a two-stage image denoising method based on an adaptive singular value threshold, which performs denoising estimation twice, and comprises the following steps of: dividing an original image to be denoised into a plurality of image blocks with the same size, and grouping similar image blocks through a block matching method to form a similar group matrix; converting a denoising problem into a low-rank approximation problem, and estimating similar image blocks of each group by using an adaptive singular value threshold; aggregating the estimated image blocks into an initial denoised image; in order to improve the first denoising result, adopting a two-stage strategy with a back projection step to further suppress noise residual errors, and generating a new noise image;and repeating the steps of the previous first stage to generate final denoising estimation. According to the method, more edge details which are easily blurred can be retained subjectively, the problem that a denoising result is too smooth is solved, the visual effect of the image is improved, and the calculation cost is reduced through fewer iterations.