SSIM and their dynamic range for image quality assessment

Recent metrics to objectively evaluate image quality based on measuring structural changes, and not on measuring the error, have led to improvements in evaluation with respect to the human vision system. Various analyses based on calculating the correlation between values obtained from subjective experiments and those obtained objectively have shown this, and variations and improvements have been gained compared to the original proposal. However, the introduction of constants to the calculations, although they improve the stability of the measurements, also reduces their dynamic range, which in turn affects the correlation with HVS. On the other hand, this metric and its improvements have also been used for video quality assessment, usually through an average of each frame. However, in many cases, the computing costs increase considerably, which limits their use in systems with high resolution and with video applications. In this work, we show how to compensate for this reduction in the dynamic range in the SSIM calculation, which could improve the correlation with the SSIM of HVS without increasing the computing cost.