Research On The Techniques Of Digital Image Inpainting

The widely-used applications of the image inpainting make this technique a hot research in the image processing techniques recently. The aim of this technique is to inpaint the damaged image in an automatic way. In such an information age of the image, more and more applications referring to image inpainting appear. The main content of this dissertation is described as follows.Firstly, digital image processing and its applications are introduced, and then the development status of the image inpainting techniques, two basic kinds of the inpainting models and the standards to evaluate the quality of the image are successively introduced.After that, we introduce an algorithm based on the curvature-driven diffusion algorithm. The curvature-driven diffusion inpainting method, which makes use of the geography information of the image to restore the missing information of the damaged image, provides a good result to restore the geography information, but the technique costs really a great deal of time, which constraints its applications. To make up for this, we notice that a simple algorithm - median filter -is a good choice to restore the pixels' value for a smooth area, so we classify the image into two kinds according to whether it is smooth or not. We adopt the method of the curvature-driven diffusion method to inpaint the damaged areas which are not smooth, and the median filter method to inpaint the left. The efficiency of this method is verified by a series of experiments.Finally, we introduce the applications of recent widely-used wavelet transformation theory applied in many fields into image inpainting technique. The proposed method models images using discrete wavelet transformation.The damaged image is decomposed into several bands of different frequencies using the DWT at first, and then sub-frequency bands are inpainted separately. The general method is used to inpaint the low frequency information. Since the high frequency contains the direction information of the edges in the image, the high frequency wavelet coefficients image is divided into many small blocks firstly, the line fitting is used to approximate the direction of the edges to each block, and then the values of the damaged pixels in the block are approximated according to this direction. Experiments show the efficiency of the method.