Full-reference image quality metric for blurry images and compressed images using hybrid dictionary learning

The image quality degradation due to the loss of high-frequency components of images is often seen in real scenarios, such as artifacts caused by image compression and image blur caused by camera shake or out of focus. Quantifying such degradation is very useful for many tasks that are related to image quality. In this paper, an effective approach is proposed for the image quality assessment on images with blur as well as images with compression artifacts. Based on the relation between the dictionaries of the degraded image and the reference image, we build up a hybrid dictionary learning model to characterize the space of patches of the reference image as well as that of the degraded image. The image quality is then measured by the difference between the two resulting dictionaries. Combined with a simple sparse-coding-based metric, the proposed method shows competitive performance on five benchmark datasets, which demonstrates its effectiveness.