Image quality metric based on regular structure features

The objective assessment of image quality has been extensively studied in recent years, and a large number of evaluation algorithms have been presented. The existing evaluation algorithms of image quality has already exhibited satisfactory performances in many common image distortion models, such as additive noise, grayscale adjustment, low pass filter, motion blurring, lossy compression, etc. However, most of them have no effect on the geometric distortion, which is widely existed in the digital images. In view of geometric distortion, a full-reference image quality assessment method is presented in this paper. The core idea of this method is that only the geometric distortion acting on regular structure features of image can have significant impacts on the perceptive quality of the image, and the physical strength and spatial irregularity of geometric distortion are the main factors causing the decline of image quality. According to experimental data, the proposed assessment method can be well applied to geometric distortion model, the evaluation results are better than most of the existing methods, and it has showed significant correlation with MOS.