最小噪声分离变换与Haar小波变换结合的壁画线状特征增强方法

在壁画的保护与修复中,线状特征具有重要的意义。然而由于自然环境等因素的影响,壁画经常出现褪色残缺等病害,导致其线状特征难以辨认。因此,利用高光谱成像与Haar小波变换结合,提出了一种壁画线状特征增强方法。首先,对高光谱影像进行最小噪声分离(MNF)变换,选取前10波段进行MNF逆变换进行重构,实现高光谱影像的降噪处理。其次,对重构后的影像选择真彩色波段变换为灰度图像,对灰度图像进行Haar小波分解。然后,对最小噪声分离变换后的影像,利用最大平均梯度法进行最优波段选择,将最优波段利用Haar小波进行变换,利用其分解后的低频信号与灰度图像分解后的低频信号相融合,高频信号使用MNF逆变换重构后的灰度图像。最后,对优化组合的低频和高频信号进行Haar小波逆变换得到结果图像,达到增强线状特征的目的。经过与原始灰度影像、主成分分析线状特征增强方法对比,验证了线状特征增强方法的有效性。研究结果可为壁画的保护修复提供更丰富、更直观的参考信息。

A method for enhancement of mural linear features: combination of minimum noise fraction and Haar wavelet transform

The study of linear features plays an important role in the field of mural protection and restoration.However,due to the natural environment or other factors,murals often suffer from many diseases which make identification of their linear features difficult.Therefore,in this study,a method combining minimum noise fraction(MNF)and Haar wavelet transform was proposed to enhance the linear features using hyperspectral images of murals.Firstly,MNF transformation was carried out on the hyperspectral images,and then the top 10 bands were selected for inverse MNF transformation for reconstruction to reduce noise of the hyperspectral image.Secondly,true color bands of the reconstructed image were transformed into a gray image,which was decomposed by Haar wavelet in the next step.Then the optimal band of the image after MNF was chosen by using the maximum average gradient algorithm.It was transformed into two parts—a low-frequency signal and high-frequency signal—by the same Haar wavelet as above.After that,this low-frequency signal part was fused with the low-frequency signal part from the gray image decomposition to get the optimized low-frequency signal.The reconstructed gray image above is considered to be the optimized high-frequency signal.Finally,the optimized low and high frequency signals were transformed inversely to the resultant image by the Haar wavelet.Comparison of the original gray image and resulting enhancement by principal component analysis,verifies the effectiveness of the linear feature enhancement method proposed.