Application Of Image Restoration Based On Regularization Methods

Image restoration is one of the basic areas of image processing. In the process of image forming, copying, scanning, transmission and displaying, the quality of image will degrade substantially, due to the imaging system, equipment or transmission media limited, as well as inevitably noise pollution. And in practice, it is even an ill posed inverse problem, which makes it unavailable to solve this kind of image restoration problem using classic direct methods. On the other hand, we should use appropriate regularization methods to solve it.This paper has deeply studied H-O model’s ideas[1] and H-O model’s "False Texture Phenomenon", as well as the reasons of this phenomenon. In order to solve H-O model’s "False Texture Phenomenon", this paper puts forward a new regularization model----Image restoration model based on low rank and high order total variation. The new model skillfully builds one regularization term that charactering texture’s low rank feature, utilizing the ideas of Maximum rank Decomposition and penalty methods. As with cartoon part, this paper uses high order total variation regularization term to constrain cartoon part, owing to high order total variation’s features that alleviating "False Texture Phenomenon" and catching cartoon. The new model this paper proposed not only restores the degraded image, but also decomposes it. So according to the new model, we obtain the restored image and its cartoon part, texture part. For the solving of the new model, this paper utilizes Augmented Lagrange method to solve it, which guarantees faster computation speed. In the end, substantial numeric experiments indicates that the new model has alleviated H-O model’s "False Texture Phenomenon" to some extent, further improved the image quality, and possesses better restoration result.