DFT-based no-reference quality assessment of blurred images

In this paper, blind image quality assessment (IQA) of Gaussian blurred images based on Discrete Fourier Transform (DFT) is proposed. The proposed work is based on the fact that if a positive constant is added to the magnitude of every AC coefficient of an image, then the average value of ratio of AC coefficient magnitude before and after the constant, gives an indication of degree of blurriness in the image. This ratio is likely to be smaller for blurred images and larger for sharper images. Further, it has been observed that the above-mentioned ratio measures the blurriness more precisely when the DFT is applied on the derivative of image instead of image itself. Therefore, in the proposed approach, the image derivative is calculated first and then block-based DFT is computed. The ratios of each AC coefficient (before and after adding a fixed positive constant) within the block are averaged at block level and then it is pooled to obtain a quality metric estimating the perceptual quality of the overall image. The proposed method is very fast and highly correlated with the subjective image quality rating and hence is suitable for real-time blurriness estimation applications.