Method Based On Hyperspectral Imaging For Enhancing Line Features In Murals

Mural as a form of painting,with a rich variety of beautiful color images,is an important part of cultural heritage,different styles of fresco reflects the cultural characteristics of different periods,fresco is not only the refinement of art and culture to the extreme embodiment,but also the organic combination of artistic beauty and social aesthetic values.But in the long years,because of environmental changes and human factors,fading disease such as bone gangrene,so that the surface of the murals become more and more "blurred",until the human eye is difficult to identify,greatly affecting the study and appreciation of the murals.In order to solve the "chronic disease" of frescoes,a fresco line feature enhancement scheme is designed to extract the faded fresco line features and combine them with image fusion methods to achieve the optimization of fresco line features.The hyperspectral technology has the characteristics of non-destructive and "spectrum-in-one",which can make full use of the rich spatial information and strong infrared penetration of hyperspectral images to record and analyze the line features on the mural more comprehensively.At the same time,relevant feature extraction methods and filtering methods are introduced to extract the line features of the mural,and further optimize the fused features by using color space conversion and wavelet transform to enhance the line feature information of the mural and improve the visual effect and artistic value of the mural.The main contents are as follows:(1)A hyperspectral line feature extraction method based on Guided Filtering(GF)and Linear Discriminant Analysis(LDA)was developed.The hyperspectral image data is collected for the areas where the ancient murals are severely faded and the line features are blurred and difficult to recognize,and the guided filtering and LDA feature extraction are used to retain the line structure information better.Firstly,the hyperspectral image after the reflectance reconstruction is extracted by using the Minimum Noise Fraction(MNF),the band with the linear features of the faded areas of the mural is used as the guide image,the feature band image with high signal-to-noise ratio is selected for MNF inversion,and the hyperspectral image after the MNF inversion is denoised is separately extracted by using the guide filter.The spatial features are extracted from each band,and the extracted spatial features are superimposed.Then,the superimposed images are divided into training samples and test samples,and the label data of different color regions are obtained through supervised classification by Support Vector Machine(SVM),and supervised LDA feature extraction is performed on the superimposed images using the label data.The best band image is selected as the line feature extraction image by comparing each band after feature extraction with the image evaluation indexes such as average gradient,information entropy and edge intensity.Finally,the line feature extraction image is fused with the true color image(hereinafter referred to as true color image)synthesized by MNF inversion,and the fused image is the line feature enhanced image.The line features in the faded areas of the line feature enhancement image are very obvious,and the line features that were previously blurred to the naked eye become very specific and clearly visible,and the results show that the method can effectively enhance the line feature information in the faded areas of the fresco pigment layer.(2)A fusion method of mural line features based on color space conversion and wavelet transform.The hyperspectral line feature extraction method based on guided filtering and LDA can effectively enhance the line features in the faded areas of the mural,but the enhanced line features have distorted colors and large noise,and the overall image quality needs to be improved.Therefore,a fusion method based on color space conversion and wavelet transform is proposed.Firstly,the true color image in RGB color space is converted to HSV color space which is easy for image processing,and the obtained V-component image and line feature extracted image are normalized to gray.Secondly,the optimal wavelet base and the optimal number of decomposition layers of the wavelet transform are determined by comparing the effects of different wavelet bases and different decomposition layers on the average gradient,information entropy and spatial frequency evaluation indexes of the fused images afterwards.Then,the line feature extraction image and V-component image are decomposed into high-frequency signal and low-frequency signal in this way,and suitable wavelet fusion rules are selected to fuse the high-frequency signal and low-frequency signal decomposed by both of them respectively to complete wavelet reconstruction.Finally,the reconstructed image is inverted from HSV color space to RGB color space instead of the original V component to obtain the linear feature fusion image.In terms of the image linear feature enhancement effect,a comprehensive evaluation and analysis using feature statistical indexes reveals that the color of the faded area enhanced linear features is more realistic,the brightness of the linear feature fusion image is more uniform,the original state of details such as cracks in the pigment layer is retained,and the overall structure of the picture is more natural and coordinated.In summary,this study achieves the enhancement of line features in the faded areas of frescoes,with significant improvement in various aspects such as color realism,clarity,accuracy,overall coordination of the picture and noise size,which can help conservators of cultural relics to obtain more accurate information on the distribution of line features of frescoes,enhance the overall visual effect and artistic value,and provide scientific reference for their conservation and restoration and research.