This paper establishes, through the principles of signal and system theory, that under the DMOS evaluation standard, image quality assessment is exclusively linked to distortion information. We introduce a novel image quality assessment model based on the Denoising Diffusion Probability Model (DDPM). This model utilizes the de noising autoencoder from DDPM for extracting image noise, coupled with a Convolutional Neural Network (CNN) for quality evaluation. By isolating distortion information, the interference from redundant information is minimized, enhancing the efficiency of model training. The model demonstrates superior performance across five datasets, comprising both synthetic and real distortions.