Wu Yadong

An Efficient Scratches Detection and Inpainting Algorithm for Old Film Restoration

This paper investigates the removal of line scratches from old films and gives a twofold contribution. First, a simple technique for detecting the scratches is proposed. This proposed method firstly applies a linear technique to enhance the scratches, and then detects the scratches using information in frames. Second, the scratch removal is approached as an inpainting problem, and it is solved by using an inpainting algorithm based on p-Laplace operator. In our experiment with old Chinese films, we show that our scheme not only obtains more accurate detection of scratches, but also gives better video quality.

Image completion algorithm based on texture synthesis

Abstract A new algorithm is proposed for completing the missing parts caused by the removal of foreground or background elements from an image of natural scenery in a visually plausible way. The major contributions of the proposed algorithm are:(1) for most natural images, there is a strong orientation of texture or color distribution. So a method is introduced to compute the main direction of the texture and complete the image by limiting the search to one direction to carry out image completion quite fast;(2) there exists a synthesis ordering for image completion. The searching order of the patches is deffned to ensure the regions with more known information and the structures should be completed beforefflling in other regions;(3) to improve the visual effect of texture synthesis, an adaptive scheme is presented to determine the size of the template window for capturing the features of various scales. A number of examples are given to demonstrate the effectiveness of the proposed algorithm.

Variational Image Deblurring Using Modified Hopfield Neural Network

We address the problem of variational image restoration by neural network in this paper. A variational image deblurring algorithm based on modified Hopfield neural network (MHNN) is proposed. Then we analyze the convergence of the deblurring algorithm and give a strategy to guarantee the algorithm to converge stable state. Comparative experimental results show that the proposed algorithm is favorable to other related algorithms

A Novel Image Restoration Algorithm Using Neural Network Based on Variational PDE Model

The variational partial differential equation (PDE) model is introduced into image restoration framework using neural network. A novel image restoration algorithm based on harmonic diffusion equation is given. And, the convergence of this restored algorithm is also analyzed. The proposed algorithm could deblur the degraded image while preserve edges. Experimental results show that the presented algorithm is superior to the traditional restoration algorithm in preserving image edges, and has a better restoration effect