|
Murals are China's precious historical and cultural heritage. They not only record the five thousand years of history of Chinese civilization, but also demonstrate the artistic creativity of the working people in ancient China. They have extremely high artistic and academic value. However, due to long-term exposure to wind and sun and man-made damage, some of the existing color murals have peeled off, color fading and other problems. In order to preserve these cultural treasures for a long time, many scholars have taken the digital restoration and protection of murals as their research direction. I regard the murals as pictures and this article will focus on the remedy of the mural color and the enhancement of the overall color, so that it can restore the original style. In the age of machine learning, I strongly recommend using machine learning for color restoration and restoration of color murals. However, unlike the traditional restoration of image color, the restoration of mural color is very complicated, and many factors need to be considered. Therefore, after referring to the method of the paper [1] of the City University of Hong Kong and Microsoft Research Asia, I optimized it and proposed a dual-domain translation network that uses real and intact color mural data and the data of various damaged color-fading murals. I train two variational autoencoders to convert them into two latent spaces respectively. The difference here with their method [1] is that the conversion between the two latent spaces is learned from real color murals and faded murals. This kind of translation has strong adaptability and integrity when used to restore the color of the mural. At the same time, in order to solve the fading problem of a mural caused by various reasons, I also referred to the methods of City University of Hong Kong and Microsoft Research Asia [1] to design a branch of global information extraction containing non-local modules. This branch is related to the fusion of latent spaces which can further improve the color recovery ability of murals that have faded due to various reasons. Especially from the results obtained later, the overall effect of this method is slightly better than other methods.
|
