Ghost Reflection Removal Using Image Gradient Sparsity Prior

When a camera is used to take an image of the opposite side through the glass,the acquired image information is usually superimposed by the transmit-ted light on the inside of the glass and the light reflected on the same side of the camera.The image information on the opposite side is captured by transmitted light,while the image information on the same side of the camera is captured by reflected light.It is assumed that the image captured consists of transmission and reflection layers.If the transmission layer and reflection layer can be sep-arated accurately to obtain the required transmission image,it will be of great aesthetic and practical value.However,the separation of the reflection and the transmission layer is an ill-posed problem.Because in the absence of other prior information,it can be decomposed in an infinite number of ways.Therefore,many researchers use different image priors or a large number of images to constrain such a highly ill-posed problem,so that the image separation with reflection is closer to the ideal result.However,these priors only apply to the scenes with a single background and obvious reflection,but not to the scenes with complex texture and more generalized.In this paper,a single image containing reflection is used as the input,and a prior reflection removal was proposed based on gradient sparsity of natural image for the situation of ghosting in the reflected image.The prior assume reflection layer and transmission layer were natural images,so it was assumed that the gradient of the reflection layer and transmission layer is sparse.Thus,the p-quasinorm(0<p<1)of the gradient between the transmission layer and the reflection layer was used as the solution of the prior term to limit the reflection separation.In the process of solving,was similar to p-norm(p?1)optimization problem,We use the Iterative weighted Least Squares(IRLS)algorithm solving p-quasinorm(0<p<1)nonconvex optimization problem to get the image of transmission and reflection layers.This paper was modeled based on the reflected image ghosting scenario.The effect of removing ghosting was more obvious for glass of a certain thickness,but it was also applicable to the case where ghosting is not obvious.We can use this algorithm to separate reflections,even though the ghosting of images are sometimes barely visible.It was found that our method was more universal and practical.Experiments on the synthetic image and the real reflection image show that our method has a significant improvement in the separation effect and calculation time compared with similar methods.