No-Reference Image Quality Assessment Based On The Statistical Characteristics Of DCT Coefficients

With the rapid development of digital image processing technology, it becomes more and more important day by day to assess the quality of a digital image accurately and effectively. Recently,image quality assessment has become a concernful subject in image information project domain.There are two representative methods mainly for image quality assessment. One is the objective assessment method; the other is the subjective assessment method. The former is comprised of full-reference quality assessment method, half-reference quality assessment method and no-reference quality assessment method. So far, most image quality assessment methods having been brought forth are full-reference or half-reference ones that require the original image as a reference. However, in practice, for most of the compressed images, the original images are not available and that is why the no-reference image quality assessment method is attached more and more importance to.A DCT Based No-Reference Digital Image Quality Assessment method is proposed in this paper, which takes the static JPEG images as study objects.It is well known that the characteristic that large amount of information is concentrated on the low frequency in the result of discrete cosine transform (DCT), which is utilized by JPEG image compression scheme, and quantization of the DCT coefficients introduces error in the image representation and causes the main loss of signal information in the whole compressing process. It is generally believed that after an image is transformed block by block (8×8) with discrete cosine transform, for each AC coefficients, the histogram of unquantized coefficients for a fixed AC frequency resembles a Laplacian distribution. This characteristic of DCT coefficients is utilized in this paper. Based on meticulous and painstaking analysis and research on this characteristic, a scheme for calculating the quantization error on DCT coefficients is proposed to be applied to digital image quality assessment. The realization process and the theoretical foundation of this scheme are explained in detail and large numbers of experimental result data are provided to prove its validity and reliability in this paper.