Optimized method for polarization-based image dehazing

Abstract

Image dehazing is desired under the foggy, rainy weather, or the underwater condition. Since the polarization-based image dehazing utilizes additional polarization information of light to de-scatter, image detail can be recovered well, but how to segment the polarization information of the background radiance and the object radiance becomes the key problem. For solving this problem, a method which combing polarization and contrast enhancement is demonstrated. This method contains two main steps, (a) by seeking the region of large mean intensity, low contrast and large mean degree of polarization, the no-object region can be found, and (b) through defining a weight function and judging whether the dehazed image can achieve high contrast and low information loss, the degree of polarization for object radiance can be estimated. Based on the estimated parameters, the scatter of light by the mediums can be diminished considerably. The theoretical derivation shows that this method can achieve advantages complementation, such as being able to obtain more details like the polarization-based method and high image contrast like the contrast enhancement based method. Besides, it is physically sound and can achieve good dehazing performance under different conditions, which has been verified by different hazing polarization images.