Cultural relic denoising method

Abstract

The invention relates to a cultural relic denoising method. Specifically, the cultural relic denoising method is a novel denoising method combining a mathematical model and a computer technology, and low-rank regularization is applied to a factor matrix of Tucker decomposition, so that the provided method avoids the limitation of an existing cultural relic denoising method, such as the problems of rank selection and low calculation efficiency. And secondly, sparse noise is removed by adopting a non-convex penalty function, and the sparse noise can be closer to a data acquisition model and a prior model, so that the damaged area of the cultural relic is complemented and repaired, and the original appearance of the cultural relic is recovered while the cultural relic is protected.

Description

Technical Field

[0001] The invention belongs to the field of artificial intelligence and pattern recognition, and specifically relates to a cultural relic denoising method. Background Art

[0002] Traditional methods for removing noise from cultural relics mainly rely on manual operations, which will inevitably have a direct impact on the cultural relics themselves during the restoration process, and may cause secondary damage. In addition, traditional methods are usually time-consuming, costly, and cannot completely restore the original appearance of the cultural relics. In this case, it becomes particularly urgent to find new, more efficient and non-destructive restoration methods.

[0003] With the rapid development of computer technology and digital technology, digital and virtual technology have provided new means for the protection and restoration of cultural relics. The development of high-precision 3D scanning technology, image processing technology, virtual reality (VR) and augmented reality (AR) technology has made it possible to digitize and denoise cultural relics. Through these technologies, cultural relics can be accurately digitally recorded and virtually reconstructed to achieve lossless restoration and permanent preservation.

[0004] The advantage of cultural relic denoising is that it simplifies and replaces complicated manual operations with digital technology, and can reduce the risks in the manual operation process. At the same time, a complete restoration plan is formulated for the objects that need to be restored, which greatly reduces the risk of denoising of physical cultural relics and realizes denoising of cultural relics in an unrestored state.

[0005] At present, the denoising methods of cultural relics are being studied in depth in terms of feature calculation, contour extraction, fracture surface matching, high-level semantic annotation, and 3D model retrieval. On this basis, a large number of experiments on splicing and reassembling Terracotta Warriors and Horses fragments have been completed by combining computer graphics, visualization, and virtual reality technology, and the virtual restoration of the Terracotta Warriors and Horses has been realized. Other researchers have used cultural relic digitization, virtual reality, and computer technology to establish a cultural relic denoising workflow for large and complex 3D models of cultural relics, and have proposed a method for predicting the size of defective parts of cultural relics based on a regression model, taking the Dazu Rock Carvings Thousand-Handed Guanyin stone cultural relic as an example.

[0006] In the present invention, a method for denoising cultural relics is proposed. Summary of the invention

[0007] The present invention relates to a method for denoising cultural relics. Specifically, a method for denoising cultural relics is a new restoration method that combines mathematical models and computer technology. By applying low-rank regularization to the factor matrix of Tucker decomposition, the proposed method avoids the limitations of existing cultural relic denoising methods, such as rank selection and low computational efficiency. Secondly, the use of a non-convex penalty function to remove sparse noise can be closer to the data acquisition model and the prior model, thereby improving the completion and restoration of damaged areas of cultural relics, while protecting the cultural relics and restoring their original appearance.

[0008] Data collection and preprocessing: Use laser scanning, structured light scanning and other equipment to perform three-dimensional scanning of cultural relics to obtain geometric data. Use high-resolution cameras to capture images of the surface of cultural relics to obtain texture information. De-noise and pre-process the acquired data to ensure data quality. Low-rank matrix decomposition: Represent three-dimensional point cloud data and image data in matrix form. Apply algorithms such as singular value decomposition (SVD) or principal component analysis (PCA) to decompose the matrix into low-rank matrices and sparse matrices. Repair and reconstruction: Fill in the damaged areas of cultural relics through reconstruction of low-rank matrices and sparse matrices. Combined with image processing technology, texture mapping is performed on the repaired geometric data to complete the color and texture information of the missing parts.

[0009] The specific steps of the above-mentioned cultural relics denoising method are as follows:

[0010] Due to the presence of objects or edges, cultural relic images usually exhibit piecewise smooth structures in spatial and spectral dimensions. It is well known that total variation regularization is an effective tool for characterizing local piecewise smooth structures and has been widely used in the field of image processing. Combined with low-rank expressions, this patent proposes a cultural relic denoising framework

[0011]

[0012] The tensor Represents the recovery result, X=g×1U1×2U2…× N U N Decomposed for Tucker, is the core tensor, is a factor matrix. Obviously, the factor nuclear norm minimization in formula (1) is a relaxed form of modulo-n rank minimization. n =I n , F nj is a matrix with dimension (In-1)xIn. Compared with the existing tensor decomposition artifact denoising method, the model proposed in this patent has a better effect on the factor matrix U n Imposing nuclear norm regularization constraints does not require a pre-given Tucker rank; at the same time, SVD calculations are performed on smaller matrices. Compared with the n-rank minimization-based method that requires SVD operation on each unfolded matrix X(n), the proposed low-rank framework can greatly reduce the computational cost, especially for large-scale data.

[0013] The advantages of this patent are as follows:

[0014] A framework for artifact denoising is proposed. In this framework, the nuclear norm is used as a low-rank regularization to avoid the difficulty of finding the optimal rank. At the same time, the nuclear norm regularization is placed on a factor matrix with a smaller dimension instead of the modulo-n expansion matrix of the original tensor, which greatly reduces the computational cost of the singular value decomposition (SVD) optimization step. Furthermore, this patent uses a non-convex and non-smooth MCP function to remove impulse noise, overcoming the disadvantage that the L1 norm may deviate from the data acquisition model and the prior model. In order to solve the proposed model, this patent designs an alternating direction multiplier method based on the symmetric Gauss-Seidel. Compared with the directly extended ADMM algorithm, the sGS-ADMM algorithm can make more full use of the structural information of the optimization problem, thereby obtaining a more accurate solution. In addition, the Difference of Convex Functions (DC) algorithm is used to solve the corresponding non-convex subproblem to avoid additional computational burden. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a comparison of the average numerical results of restoration of 8 cultural relics images with different methods and different noise levels. DETAILED DESCRIPTION

[0016] The present invention will be further described below in conjunction with embodiments:

[0017] Example 1

[0018] Figure 1 The figure shows the average numerical results of the restoration of 8 cultural relic images by FTDTV-MCP and 7 comparison methods at each noise level, where the best results are shown in bold. It is not difficult to find that in all experiments, FTDTVMCP achieved the highest PSNR, SSIM and lowest RSE. In order to further compare the restoration performance of all methods, thanks to the protection of detail information and edge structure by L0 gradient regularization and spatial-spectral total variation regularization, LRTFL0, LRTDTV and FTDTV-MCP proposed in this patent have better visual effects than the other five methods, especially for images with rich details, whose contours and textures are clearer and closer to real images. Compared with LRTFL0 and LRTDTV, the improvement in PSNR value of the method proposed in this patent further demonstrates the effectiveness of low-rank factor prior and non-convex sparse regularization.

Claims

1. A method for denoising cultural relics, characterized in that: The steps include: (I) Data collection and preprocessing: Use laser scanning, structured light scanning and other equipment to perform three-dimensional scanning of cultural relics to obtain geometric data. Use high-resolution cameras to capture images of the surface of cultural relics to obtain texture information. De-noise and preprocess the acquired data to ensure data quality. (II) Low-rank matrix decomposition: Represent the 3D point cloud data and image data in matrix form. Apply algorithms such as singular value decomposition (SVD) or principal component analysis (PCA) to decompose the matrix into low-rank matrices and sparse matrices. Repair and reconstruction: fill in the damaged areas of the cultural relics through reconstruction of low-rank matrices and sparse matrices. Combined with image processing technology, texture mapping is performed on the repaired geometric data to complete the color and texture information of the missing parts. The specific steps are as follows: Due to the presence of objects or edges, cultural relic images usually exhibit piecewise smooth structures in spatial and spectral dimensions. It is well known that total variation regularization is an effective tool for characterizing local piecewise smooth structures and has been widely used in the field of image processing. Combined with low-rank expressions, this patent proposes a cultural relic denoising framework The tensor Represents the recovery result, X=g×1U1×2U2…× N U N Decomposed for Tucker, is the core tensor, is a factor matrix. Obviously, the factor nuclear norm minimization in formula (1) is a relaxed form of modulo-n rank minimization. n =I n , F nj It is a matrix of dimension (In-1)xIn. Compared with the existing tensor decomposition artifact denoising method, the model proposed in this patent has a better effect on the factor matrix U n Imposing nuclear norm regularization constraints does not require a pre-given Tucker rank; at the same time, SVD calculations are performed on smaller matrices. Compared with the n-rank minimization-based method that requires SVD operation on each unfolded matrix X(n), the proposed low-rank framework can greatly reduce the computational cost, especially for large-scale data.

Images