A New Algorithm For Integrated Image Quality Measurement Base On Wavelet Transform And Human Visual System

Using the properties of human visual system (HVS), the combination of subjective and objective measurement methods is the most effective way to assess image quality, and also the trend of this research field. According to this idea and taking advantage of the similarities of discrete wavelet transform (DWT) with HVS, a new algorithm for integrated image quality assessment is presented in this paper.Firstly, the mutual relationship between the DWT and HVS is analyzed, which becomes the theory basis of the new algorithm. An important question that has no conclusion yet, the selection of wavelet bases in the practical application, is also discussed and summarized, and the fit wavelet base function of the new algorithm is selected. Based on DWT, we make lots of singleband noise images corresponding to different wavelet subbands, and establish the relationship between the subjective and objective assessment results of noise images after series of experiments and analysis. The new algorithm is composed of wavelet Weighted mean square error (WWMSE) and image quality scale (IQS) choosing face images as research object. Adopting CCIR 500 recommendation standard and the peak mean square error (PMSE) criterion, we obtain the CCIR five-point noise thresholds of each subband corresponding to four-level wavelet decomposition. Combining the properties of HVS to analyze the sensitivity to each subband's noise thresholds, we thenobtain the visual weighted coefficient of each subband. Thus the new method for image quality objective measure, WWMSE, is proposed by having visual weighted processing to the PMSE of singleband target image. According to these, we finally establish mathematical models of HVS noise threshold and IQS based on CCIR five-point quality scale. Assigning value to the model parameters, we can measure the IQS of assessed target image directly, which replace the traditional method of mean opinion score (MOS) effectively. In this paper, besides the systemic theory research, we also do lots of emulation experiments especially the obtaining of noise thresholds and weighted coefficients of subbands, that founds the credible basis for the new algorithm we proposed in this paper.In terms of experiment results, the new algorithm shows good feasibility, reliability and validity comparing with conventional image quality assessment methods. It not only preserves the merits of conventional image quality measurement methods, but also overcome their defects such as time-consuming, costly and objective assessment results not accordance with subjective perception quality, etc., effectively. The new algorithm we proposed in this paper reliably realizes the goal of the combination of subjective and objective measurement methods, and also provides an effective analysis method for modern image processing technology.