An Improved Total Variation Model For Image Inpainting

Image inpainting technology is an important part of image processing.The technology is widely used in the removal of text and scratches in images.Its main work is to fill the damaged area of the image according to certain rules,resulting in inpainting its original image.The methods of image inpainting can be classified into three ways according to the size of the damaged area.They are image structure-based inpainting method,image texture-based inpainting method and image decomposition-based inpainting method.The paper focuses on the image structure-based inpainting method.This method usually uses partial differential equation as a tool.It is suitable for inpainting images with small damaged area.The Total Variation(TV)model is a classical model in this method.After that,the model has been improved by many scholars,whereas there still exists some defects.In view of the unnatural phenomenon of gray level transition in the inpainted images of existing TV model,this paper introduces the image enhancement terms of Snake model,and proposes a new model.Furthermore,the inpainting experiments of damaged green pepper images verify that the new model can enhance the visual connectivity of the image and make the gray level transition more natural.Because of the existing TV model and the new model making use of the four neighborhood information of the inpainted points in the numerical solution,it is poor in image inpainting accuracy.Thus,the double-cross algorithm based on the improved TV model and the double cross algorithm based on the new model are further used.After that,the improved algorithm can improve the accuracy of the image by inpainting the damaged Lena image and damaged animation image.