Image Inpainting Based On Partial Differential Equation

It is a common image inpainting method to utilize partial differential equation (PDE). At present, digital image inpainting technique has been widely used for removing occluding objects and filling in damaged portion of images. Image inpainting can be used to meet some special demands.Firstly, this paper introduced advantage of digital image, meaning of digital image inpainting, development of digital image inpainting; secondly, discussed algorithm for image inpainting based on BSCB(Bertalmio-Sapiro-Caselles-Ballester) model; This was a successful example of image inpainting based on partial differential equation. The procedure by BSCB model is similar to manual processing. The method can keeps complete edges, but it does not meet connectivity principle. When the area to be restored is narrow, the result is good. The complexity of BSCB model is higher than other models. Next, this paper discussed in detail the application of TV model in image inpainting. The model is based on PDE. While image being restored, the effect of noise is removed. The complexity of TV model is lower than BSCB model. It does not meet connectivity principle also.Chapter 5 is the important portion of this paper. In this chapter, studied chiefly digital mage inpainting based on improved heat transfer model. Commonly, the equation presented heat transfer is a two order PDE. The isotropy model can be applied to image inpainting, but the results are not very good. The processing spends a long time. Two algorithms that based on improved heat transfer model have been presented in this paper. One of them considers specific position. When inpainting is processing, the method is special (anisotropy method). The other considers correlated relation of neighborhood pixel, the energy of digital image that was restored is the lowest. This paper defined the norm of complex vector. The digital image may be restored by the algorithm based on a renovated heat conduction model. Complex vector has been proposed. The two methods have been contrasted with other algorithm. The experiment results suggested the improved heat transfer model would spend less time than other model. Especially, the algorithm based on neighboring 1 norm got more reasonable result and more fast speed.