6533b85dfe1ef96bd12beff5

RESEARCH PRODUCT

The impact of noise estimation on dehazing

subject

description

International audience; Under haze or fog, the quality of the images is degraded due to the atmosphere, causing the details of the images to be difficult to identify by observers and computer vision systems. Such images contain noise, which is mainly due either to environment (extrinsic noise) or sensor (intrinsic noise). As the transmission of light coming from the scenes' objects is exponentially attenuated and comes quickly down to zero in presence of haze, the noise is greatly amplified at high haze densities and long distances. In order to investigate the importance of the accurate estimation and the removal of noise from hazy images, we used the CHIC (Color Hazy Image for Comparison) database, which provides, for a given scene, the haze-free image and a set of images with different haze densities. For each scene, several parameters are available like the distance from the camera of known objects such as Macbeth Color Checkers, their radiance, and the haze level through transmittance. At two levels of haze, we added some Gaussian noise. We first applied dehazing without considering the induced noise. Later, we applied it with including the accurate value of added noise and finally by using biased values. This study shows the importance to estimate as accurately as possible the noise in order to remove it and guarantee a high quality and a good recovery of images' features after dehazing.