Preferences for the balance between true image detail and noise
Sachin Gangadhar Deshpande, Scott J. Daly · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2011
We designed a series of experiments to measure user preference for the noise-detail tradeoff, including tests of the assumption that all true image detail is preferred. We generated samples with noise-detail tradeoff by designing a sequence of coring filters with increasing strength. A user study method is developed using magnitude estimation approach. In the first experiment the coring filter sequence is applied to original video samples without any additional noise. It is observed that the subjective quality score increases as coring strength is increased, reaches a peak and then decreases. Thus users prefer slightly cored images compared to original images. In the second experiment the coring filter sequence is applied to video samples with additive noise of different strength. It is observed that the most preferred coring strength increases as the amount of noise in the image increases. The results from our experiments can be used to design parameters for various image/ video post-processing and noise removal algorithms.