The invention discloses a multi-resolution JND (Just Noticeable Difference) model building method based on visual perception. The method comprises the steps of: 1. arranging a group of balanced multi-wavelet filters, and carrying out multi-channel wavelet decomposition on an input image through the group of balanced multi-wavelet filters; 2. respectively calculating a basic JND threshold, a luminance masking factor and a contrast masking factor of a luminance layer and a chroma layer, so as to simulate a space contrast sensitive function effect, a luminance self-adaptive masking effect and a contrast masking effect; and 3. respectively calculating to obtain a multi-resolution JND model based on visual perception by utilizing the luminance layer and the chroma layer. According to the multi-resolution JND model building method based on the visual perception, the characteristics of the human visual system are effectively simulated, and the multi-channel decomposition, the space contrast sensitive function effect, the luminance self-adaptive masking effect, the contrast masking effect and the like are compressively taken into account, so that the multi-resolution JND model based on the visual perception can be built; and the model can inject multiple noises, and also can keep the image to be in high visual quality after the noises are injected.