Evaluation and Improvement of Noise Estimation from a Single Image

Kenji Kamimura, Hitomi Ito, Norimichi Tsumura, Toshiya Nakaguchi and Yoichi Miyake

Graduate School of Advanced Integrated Science, Chiba University, Inage-ku, Chiba, Japan

The Journal of the institute of image information and television engineers (Accepted to be published in Vol.63, No.1, pp.101-104, 2009 ).

Abstract
The estimation of noise level of a digital image is important for many applications such as noise removal, image quality assessment and so on. From the aspect of practical use in these applications, the estimation should be achieved from a single image. Ce et al. proposed a single image noise estimation. In their method, a priori information of noise characteristics was generated by the computer simulation and used in Bayesian MAP. However, they did not consider inherent characteristics of each camera and that depress the estimation accuracy. Therefore in this paper, we take into account inherent characteristics of camera by using camera specific a priori in Bayesian MAP. Camera specific a priori is constructed from preliminarily measurement of camera instead of computer simulation. The experimental results show the improved accuracy of noise estimation.

Full paper ( to appeare )
[PDF] (in Japanese])


Back