WO2023078285A1 - 一种文字图像的摩尔纹祛除方法、装置及电子设备 - Google Patents

一种文字图像的摩尔纹祛除方法、装置及电子设备 Download PDF

Info

Publication number
WO2023078285A1
WO2023078285A1 PCT/CN2022/129187 CN2022129187W WO2023078285A1 WO 2023078285 A1 WO2023078285 A1 WO 2023078285A1 CN 2022129187 W CN2022129187 W CN 2022129187W WO 2023078285 A1 WO2023078285 A1 WO 2023078285A1
Authority
WO
WIPO (PCT)
Prior art keywords
matrix
grayscale image
reconstructed
image
energy spectrum
Prior art date
Application number
PCT/CN2022/129187
Other languages
English (en)
French (fr)
Inventor
邓元策
Original Assignee
瞬联软件科技(北京)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 瞬联软件科技(北京)有限公司 filed Critical 瞬联软件科技(北京)有限公司
Publication of WO2023078285A1 publication Critical patent/WO2023078285A1/zh

Links

Images

Classifications

    • G06T5/70
    • G06T5/92
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10004Still image; Photographic image
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20048Transform domain processing

Definitions

  • the invention relates to a method for removing moiré patterns of text images, and also relates to a corresponding device for removing moiré patterns and electronic equipment, belonging to the technical field of image processing.
  • moiré is the result of interference caused by the difference in refresh rates between the smartphone's image sensor and the display screen. Moiré will seriously affect the quality of captured images, and even cause their information to be lost and unrecognizable. Therefore, in recent years, there have been many related researches and technical solutions to reduce or eliminate moiré.
  • the primary technical problem to be solved by the present invention is to provide a method for removing moiré in text images, which can reduce or even eliminate moiré in text images.
  • Another technical problem to be solved by the present invention is to provide a moiré removal device for text images.
  • Another technical problem to be solved by the present invention is to provide a corresponding electronic device.
  • a method for removing moiré in text images including the following steps:
  • the energy prominent region includes the first energy prominent region caused by the high frequency lines of the moiré pattern and The second energy prominent area caused by the inherent texture of the text;
  • the reconstructed spectral matrix E is sequentially subjected to spectral translation and inverse Fourier transform to obtain a reconstructed grayscale image ⁇ .
  • the pair of the translation spectrum matrix Perform binarization processing to obtain the energy spectrum mask matrix Z, which specifically includes:
  • a threshold value decision is made on the energy spectrum matrix X by the following formula to obtain a binarized energy spectrum mask matrix Z, namely:
  • z (i, j) and x (i, j) are the elements of the energy spectrum mask matrix Z and the energy spectrum matrix X in row i and column j, respectively, and T is the decision threshold, which is obtained by overall balancing the image brightness.
  • the coordinates of the elements contained in each of the connected regions and the number of elements in all connected regions are obtained, and exponential filtering is performed on all non-maximum connected regions to obtain a filtered new energy spectrum mask matrix ⁇ , which specifically includes :
  • said estimating the center coordinates of any non-maximally connected region specifically includes:
  • the following cost function is iterated by the clustering algorithm to estimate the center coordinate C ⁇ (x o , y o ) of any non-maximally connected region S ⁇ ;
  • is the jth element in the connected region element set S ⁇ , and its coordinates are (x j ,y j ); C ⁇ (x o , y o ) is the central coordinate of the connected region; the non-maximally connected region S ⁇ Contains a total of H elements; ⁇ * ⁇ is an operator for distance measurement, including at least Euclidean distance or Mahalanobis distance operation.
  • the traversal of all element coordinates in the non-maximum connected region performs exponential filtering according to the distance between the coordinates of each element in the non-maximally connected region and the center coordinate, so as to complete a non-maximum connected region
  • Exponential filtering of connected regions including:
  • the reconstructed spectral matrix E is sequentially subjected to spectral translation and inverse Fourier transform to obtain a reconstructed grayscale image Specifically include:
  • the spectral matrix after the translation is described by the following formula Perform a two-dimensional inverse Fourier transform to obtain the reconstructed grayscale image ⁇ ;
  • the moiré removal method also includes the following steps:
  • the non-linear stretching transformation is performed on the reconstructed grayscale image ⁇ to obtain the final processed image, which specifically includes:
  • the reconstructed grayscale image ⁇ is iterated by the following formula to complete the nonlinear stretching transformation of the reconstructed grayscale image ⁇ ;
  • ⁇ i+1 ⁇ i +(1- ⁇ i ) ⁇ i * ⁇
  • ⁇ i +1 is the output grayscale image of the ith iteration
  • ⁇ i is the input grayscale image of the ith iteration
  • is the adjustment sensitivity parameter.
  • a moiré removal device including a controller, and further including:
  • An image conversion unit connected to the controller, for converting the text image into a grayscale image, and normalizing the grayscale image to obtain a grayscale image matrix D;
  • the first matrix processing unit connected to the controller, is used to sequentially perform two-dimensional Fourier transform, spectral shift and modulo operation on the grayscale image matrix D to obtain a shifted spectral matrix
  • a binarization processing unit connected to the controller, for processing the translated spectral matrix Perform binarization processing to obtain the energy spectrum mask matrix Z;
  • a connected area marking unit connected to the controller, for performing a morphological closing operation on the energy spectrum mask matrix Z, and marking the connected areas;
  • An exponential filtering unit connected to the controller, used to obtain the coordinates of the elements contained in each connected region and the number of elements in all connected regions, perform exponential filtering on all non-maximum connected regions, and obtain a filtered new energy spectrum mask membrane matrix ⁇ ;
  • a matrix reconstruction unit connected to the controller, for combining the new energy spectrum mask matrix ⁇ with the translation spectrum matrix Perform dot multiplication to obtain the reconstructed spectral matrix E;
  • the second matrix processing unit connected to the controller, is used to sequentially perform spectrum translation and inverse Fourier transform on the reconstructed spectrum matrix E to obtain a reconstructed grayscale image ⁇ ;
  • the nonlinear stretching unit is connected with the controller, and is used for performing nonlinear stretching transformation on the reconstructed grayscale image ⁇ to obtain a final processed image.
  • an electronic device which includes:
  • a memory, the program or instruction of the moiré removal method is stored in the memory
  • the processor is coupled with the memory, and is configured to execute programs or instructions in the memory, so that the electronic device executes the moiré removal method.
  • the moiré removal method provided by the present invention, by analyzing the energy spectrum of the text image, marks the connected regions of the energy binary image, and performs exponential filtering on the non-maximum connected regions, calculates the mask representation of the energy spectrum, and finally passes the spectral matrix
  • the reconstruction weakens or even completely eliminates the moiré pattern from the text image. This method does not depend on the training data set and has good robustness; compared with algorithms based on deep learning networks, the amount of calculation is greatly reduced, and it is especially suitable for embedded devices such as mobile phones.
  • Fig. 1 is a schematic flow chart of a moiré removal method provided by an embodiment of the present invention
  • FIG. 2 is a schematic diagram of the original spectral matrix Y in an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of a filtered new energy spectrum mask matrix ⁇ in an embodiment of the present invention.
  • FIG. 4 is a schematic diagram of a text image before processing in an embodiment of the present invention.
  • FIG. 5 is a schematic diagram of a processed text image in an embodiment of the present invention.
  • FIG. 6 is a schematic structural diagram of a moiré removal device provided by an embodiment of the present invention.
  • FIG. 7 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention.
  • a moiré removal method for text images provided by an embodiment of the present invention at least includes the following steps:
  • the text image needs to be preprocessed, and the steps of the preprocessing are: first, the text image is converted into a grayscale image, and then the grayscale image is Perform normalization.
  • the purpose of normalization is to avoid data overflow in subsequent operations, and the process of normalization processing is a conventional technical means, which will not be repeated here.
  • steps S21-S22 are included:
  • the grayscale image matrix D after normalization is set to be a two-dimensional matrix of m ⁇ n, and the grayscale image matrix D is subjected to two-dimensional Fourier transform according to the following formula: Transform to get the original spectral matrix Y:
  • i is the imaginary unit
  • d (j, k) is the element in row j and column k in matrix D
  • y (p, q) is the element in row p and column q of the original spectral matrix Y.
  • the first quadrant, the second quadrant, the third quadrant and the fourth quadrant are divided with the intersection point O of half of the row and column of the spectral matrix, that is, the first and second quadrants in Figure 2 , 3, and 4 show the position. It can be understood that, in the original spectral matrix Y, the region near the origin O is a high-frequency region, and the positions at the four corners of the original spectral matrix Y are low-frequency regions.
  • the first quadrant and the third quadrant of the original spectral matrix Y are exchanged, and the second quadrant and the fourth quadrant of the original spectral matrix Y are exchanged to obtain the translated translation spectrum matrix
  • steps S31-S32 are included:
  • z (i, j) and x (i, j) are the elements of the energy spectrum mask matrix Z and the energy spectrum matrix X in row i and column j respectively, and T is the decision threshold, which is obtained by overall balancing the image brightness, For example: through brightness accumulation histogram or by otsu method.
  • the energy-prominent regions include the first energy-prominent regions caused by the high-frequency lines of moiré patterns and the first energy-prominent regions caused by the text
  • the second energy highlights the area caused by the inherent texture.
  • steps S41-S42 are included:
  • S41 Perform a morphological closing operation on the energy spectrum mask matrix Z using a disc structure, wherein the disc structure has a set radius.
  • the radius of the disc structure is recommended to be 8, and of course it can be other values. Therefore, by performing a morphological closing operation on the energy spectrum mask matrix Z, the "small energy islands” in the energy prominent area can be removed, thereby reducing the first energy caused by the "small energy islands” to the subsequent moiré high-frequency lines Highlight area effects.
  • the "small energy island” refers to: a certain high-energy point, but a small and scattered area.
  • S42 According to the morphological closing operation, mark the connected regions of the energy spectrum mask matrix Z through the Fanhong filling algorithm.
  • the Fanhong filling algorithm is a conventional algorithm known at present, and will not be repeated here. Therefore, by marking the connected regions of the energy spectrum mask matrix Z, the positions of the first energy-prominent region and the second energy-prominent region can be clearly distinguished, so as to facilitate subsequent energy weakening of the first energy-prominent region.
  • the most connected region is the most important part of the foreground of the entire image, if it is processed, it may cause a large change in the brightness and color difference of the entire image. Therefore, in order to avoid too much impact on the image, the largest connected region is not processed, and only exponential filtering is performed on all non-maximum connected regions.
  • steps S51-S53 are included:
  • the following cost function is iterated through the clustering algorithm to estimate the center coordinate C ⁇ (x o , y o ) of any non-maximally connected region S ⁇ .
  • is the jth element in the element set S ⁇ of the connected region, and its coordinate is (x j , y j ); C ⁇ (x o , y o ) is the center coordinate of the connected region; Contains H elements; ⁇ * ⁇ is an operator for distance measurement, including but not limited to Euclidean distance, Mahalanobis distance, etc.
  • S52 traverse all the element coordinates in the non-maximally connected region, and perform exponential filtering according to the distance between the coordinates of each element in the non-maximally connected region and the center coordinate, so as to complete the exponential filtering of a non-maximally connected region;
  • steps S51-S52 are repeated until the exponential filtering is completed on all non-maximally connected regions, so as to obtain the filtered new energy spectrum mask matrix ⁇ .
  • steps S51-S52 are repeated until the exponential filtering is completed on all non-maximally connected regions, so as to obtain the filtered new energy spectrum mask matrix ⁇ .
  • the numerical values in FIG. 3 are only examples, and do not constitute data limitations for the new energy spectrum mask matrix ⁇ .
  • is a dot multiplication operation, that is, two matrix elements are multiplied separately. Thereby, the moiré high-frequency line interference of the text image can be eliminated.
  • steps S71-S72 are included:
  • the first quadrant of the reconstructed spectral matrix E is exchanged with the third quadrant, and the second quadrant of the reconstructed spectral matrix E is exchanged with the fourth quadrant to obtain the translated spectral matrix
  • the processed image can be re-transformed back to a grayscale image.
  • the reconstructed grayscale image ⁇ may have uneven brightness, so the reconstructed grayscale image ⁇ is iterated by the following formula to obtain Complete the nonlinear stretching transformation of the reconstructed grayscale image ⁇ ;
  • ⁇ i+1 ⁇ i +(1- ⁇ i ) ⁇ i * ⁇
  • ⁇ i +1 is the output grayscale image of the ith iteration
  • ⁇ i is the input grayscale image of the ith iteration
  • is the adjustment sensitivity parameter.
  • Figure 4 is a schematic diagram of the text image before bit processing
  • Figure 5 is a schematic diagram of the text image after steps S1-S8 are processed. Compare Figure 4 and Figure 5 It can be seen that the moiré pattern of the text image is weakened, and the text part is relatively intact, without causing obvious damage to the text information.
  • the moiré removal method marks the connected regions of the energy binary image by analyzing the energy spectrum of the text image, and performs exponential filtering on the non-maximum connected regions to calculate the mask of the energy spectrum.
  • the film indicates that moiré will be weakened or even completely eliminated from the text image through the reconstruction of the spectral matrix.
  • This method does not depend on the training data set and has good robustness; compared with the algorithm based on deep learning network, the calculation amount is greatly reduced, and it is suitable for embedded devices such as mobile phones.
  • the embodiment of the present invention also provides a moiré removal device, at least comprising: an image conversion unit 1, a first matrix processing unit 2, a binarization processing unit 3, a connected region marking unit 4, and an exponential filtering unit 5.
  • a matrix reconstruction unit 6 a second matrix processing unit 7 , a nonlinear stretching unit 8 and a controller 9 .
  • the image conversion unit 1 is connected with the controller 9, and is used to convert the text image into a grayscale image, and normalize the grayscale image to obtain a grayscale image matrix D; the first matrix processing unit 2 and the control Connected to the device 9, used to sequentially perform two-dimensional Fourier transform, spectral shift and modulo operation on the grayscale image matrix D, to obtain the shifted spectral matrix
  • the binarization processing unit 3 is connected with the controller 9 for the spectral matrix after translation Carry out binarization processing to obtain the energy spectrum mask matrix Z;
  • the connected area marking unit 4 is connected to the controller 9, and is used to perform morphological closing operation on the energy spectrum mask matrix Z, and mark the connected areas;
  • the exponential filter unit 5 is connected with the controller 9, used to obtain the coordinates of the elements contained in each connected region and the number of elements in all connected regions, perform exponential filtering on all non-maximum connected regions, and obtain the filtered new energy spectrum mask matrix ⁇ ; matrix reconstruction
  • the unit 6 is connected with the controller 9, and is used to
  • An embodiment of the present invention also provides an electronic device, including at least a memory and a processor.
  • the program or instruction of the method for eliminating moiré is stored in the memory; the processor is coupled to the memory for executing the program or instruction in the memory, so that the electronic device executes the method for eliminating moiré in the first embodiment above.
  • the electronic devices mentioned here refer to computer devices that can be used in a mobile environment and support GSM, EDGE, TD_SCDMA, TDD_LTE, FDD_LTE and other communication standards, including smartphones, notebook computers, tablet computers, vehicle-mounted computers, etc.
  • the electronic device includes at least a processor and a memory, and may further include a communication component, a sensor component, a power supply component, a multimedia component, and an input/output interface according to actual needs.
  • a communication component a sensor component
  • a power supply component a multimedia component
  • an input/output interface a multimedia component
  • memory, communication components, sensor components, power supply components, multimedia components and input/output interfaces are all connected with the processor.
  • the memory can be Static Random Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic memory, flash memory, etc.
  • the processor can be a central processing unit (CPU), a graphics processing unit (GPU), a field programmable logic gate array (FPGA), an application-specific integrated circuit (ASIC), a digital signal processing ( DSP) chips, etc.
  • CPU central processing unit
  • GPU graphics processing unit
  • FPGA field programmable logic gate array
  • ASIC application-specific integrated circuit
  • DSP digital signal processing

Abstract

本发明公开了一种文字图像的摩尔纹祛除方法、装置及电子设备。该方法包括如下步骤:将文字图像转化为灰度图像并归一化处理,得到灰度图像矩阵;将灰度图像矩阵依次二维傅里叶变换和频谱平移,得到平移谱矩阵;对平移谱矩阵二值化处理,得到能量谱掩膜矩阵;对能量谱掩膜矩阵进行形态学闭运算,并标记连通区域;获取所有连通区域的坐标和数量,对所有非最大连通区域进行指数滤波,得到新能量谱掩膜矩阵;将新能量谱掩膜矩阵与平移谱矩阵点乘,得到重建谱矩阵;将重建谱矩阵依次进行频谱平移和逆傅里叶变换,得到重建灰度图像;对重建灰度图像进行非线性拉伸变换,得到最终处理图像。该方法不依赖训练数据集,可以减少计算量并有效保护文字信息。

Description

一种文字图像的摩尔纹祛除方法、装置及电子设备 技术领域
本发明涉及一种文字图像的摩尔纹祛除方法,同时也涉及相应的摩尔纹祛除装置及电子设备,属于图像处理技术领域。
背景技术
在工作和生活中,智能手机使我们更加方便的记录和分享重要的信息。虽然智能手机的拍摄质量不断提高,但是在拍摄显示器屏幕的时候,仍然不可避免地出现奇怪的纹路。这种纹路被称之为摩尔纹。
从本质上说,摩尔纹是由于智能手机的图像传感器和显示器屏幕的刷新率不同而形成的干扰结果。摩尔纹会严重影响拍摄图像的质量,甚至让其信息丢失,无法识别。因此,近年来出现了不少的相关研究和技术方案,用来减弱或者消除摩尔纹。
发明内容
本发明所要解决的首要技术问题在于提供一种文字图像的摩尔纹祛除方法,可以减少甚至祛除文字图像中的摩尔纹。
本发明所要解决的另一技术问题在于提供一种文字图像的摩尔纹祛除装置。
本发明所要解决的另一技术问题在于提供一种相应的电子设备。
为了实现上述目的,本发明采用以下的技术方案:
根据本发明实施例的第一方面,提供一种文字图像的摩尔纹祛除方法,包括以下步骤:
将文字图像转化为灰度图像,并对所述灰度图像进行归一化处理,得到灰度图像矩阵D;
将所述灰度图像矩阵D依次进行二维傅里叶变换、频谱平移和取模运算,得到平移后的谱矩阵
Figure PCTCN2022129187-appb-000001
对所述平移后的谱矩阵
Figure PCTCN2022129187-appb-000002
进行二值化处理,得到能量谱掩膜矩阵Z,以确认能量谱掩膜矩阵Z中的能量突出区域,所述能量突出区域包括因摩尔纹的高频纹路而导致的第一能量突出区域和因文字固有纹路而导致的第二能量突出区域;
对所述能量谱掩膜矩阵Z进行形态学闭运算,并对连通区域进行标记,以对所述能量突出区域进行能量优化,并区分所述第一能量突出区域与所述第二能量突出区域的位置;
获取各所述连通区域所包含元素的坐标以及所有连通区域的元素数量,对所有非最大连通区域进行指数滤波,得到滤波后的新能量谱掩膜矩阵θ,以削弱或消除所述第一能量突出区域的能量参数;
将所述新能量谱掩膜矩阵θ与所述平移谱矩阵
Figure PCTCN2022129187-appb-000003
进行点乘运算,得到重建谱矩阵E,以消除文字图像的摩尔纹高频纹路干扰;
将所述重建谱矩阵E依次进行频谱平移和逆傅里叶变换,得到重建后的灰度图像Ψ。
其中较优地,所述对所述平移谱矩阵
Figure PCTCN2022129187-appb-000004
进行二值化处理,得到能量谱掩膜矩阵Z,具体包括:
取出所述平移谱矩阵
Figure PCTCN2022129187-appb-000005
中各个元素的模,组成同维度的能量谱矩阵X;
通过下列公式对所述能量谱矩阵X做阈值判决,得到二值化的能量谱掩膜矩阵Z,即:
Figure PCTCN2022129187-appb-000006
其中,z (i,j)和x(i,j)分别是能量谱掩膜矩阵Z和能量谱矩阵X在i行j列的元素,T为判决阈值,通过对图像亮度进行整体平衡得到。
其中较优地,所述获取各所述连通区域所包含元素的坐标以及所有连通区域的元素数量,对所有非最大连通区域进行指数滤波,得到滤波后的新能量谱掩膜矩阵θ,具体包括:
估算任意一个非最大连通区域的中心坐标;
遍历所述非最大连通区域中的所有的元素坐标,根据所述非最大连通区域中各元素坐标与所述中心坐标之间的距离进行指数滤波,以完成一个非最大连通区域的指数滤波;
遍历所有非最大连通区域,以通过相同的方式完成所有非最大连通区域的指数滤波,得到滤波后的新能量谱掩膜矩阵θ。
其中较优地,所述估算任意一个非最大连通区域的中心坐标,具体包括:
通过聚类算法对如下代价函数进行迭代,以估算出任意一个非最大 连通区域S τ的中心坐标C τ(x o,y o);
Figure PCTCN2022129187-appb-000007
其中,
Figure PCTCN2022129187-appb-000008
为连通区域元素集合S τ中的第j个元素,其坐标为(x j,y j);C τ(x o,y o)为该连通区域的中心坐标;所述非最大连通区域S τ内共包含H个元素;‖*‖为距离度量的运算符,至少包括欧氏距离或马氏距离运算。
其中较优地,所述遍历所述非最大连通区域中的所有的元素坐标,根据所述非最大连通区域中各元素坐标与所述中心坐标之间的距离进行指数滤波,以完成一个非最大连通区域的指数滤波,具体包括:
遍历所述非最大连通区域S τ中的所有的元素坐标,j=0,...,H-1,按照下式进行指数滤波;
Figure PCTCN2022129187-appb-000009
其中,α=range(x∈S τ),β=range(y∈S τ),x∈S τ和y∈S τ分别表示连通区域元素集合S τ中所有元素的x坐标集合和y坐标集合,函数range(w)=max(w)-min(w)。
其中较优地,所述将所述重建谱矩阵E依次进行频谱平移和逆傅里叶变换,得到重建后的灰度图像
Figure PCTCN2022129187-appb-000010
具体包括:
将所述重建谱矩阵E的第一象限与第三象限交换,并将所述重建谱矩阵E的第二象限与第四象限交换,得到平移后的谱矩阵
Figure PCTCN2022129187-appb-000011
通过下列公式对所述平移后的谱矩阵
Figure PCTCN2022129187-appb-000012
进行二维逆傅里叶变换,得到重建后的灰度图像Ψ;
Figure PCTCN2022129187-appb-000013
其中,
Figure PCTCN2022129187-appb-000014
i为虚数单位,
Figure PCTCN2022129187-appb-000015
为矩阵
Figure PCTCN2022129187-appb-000016
中j行k列的元素,ψ (p,q)为灰度图像Ψ的p行q列的元素。
其中较优地,所述摩尔纹祛除方法还包括以下步骤:
对所述重建后的灰度图像Ψ进行非线性拉伸变换,得到最终的处理图 像。
其中较优地,所述对所述重建后的灰度图像Ψ进行非线性拉伸变换,得到最终的处理图像,具体包括:
通过下列公式对所述重建后的灰度图像Ψ进行迭代,以完成对所述重建后的灰度图像Ψ的非线性拉伸变换;
Ψ i+1=Ψ i+(1-Ψ i)⊙Ψ i*∈
其中,Ψ i+1为第i次迭代的输出灰度图像,Ψ i第i次迭代的输入灰度图像,∈为调节灵敏度参数。
根据本发明实施例的第二方面,提供一种摩尔纹祛除装置,包括控制器,还包括:
图像转化单元,与控制器连接,用于将文字图像转化为灰度图像,并对所述灰度图像进行归一化处理,得到灰度图像矩阵D;
第一矩阵处理单元,与所述控制器连接,用于对所述灰度图像矩阵D依次进行二维傅里叶变换、频谱平移和取模运算,得到平移后的谱矩阵
Figure PCTCN2022129187-appb-000017
二值化处理单元,与所述控制器连接,用于对所述平移后的谱矩阵
Figure PCTCN2022129187-appb-000018
进行二值化处理,得到能量谱掩膜矩阵Z;
连通区域标记单元,与所述控制器连接,用于对所述能量谱掩膜矩阵Z进行形态学闭运算,并对连通区域进行标记;
指数滤波单元,与所述控制器连接,用于获取各所述连通区域所包含元素的坐标以及所有连通区域的元素数量,对所有非最大连通区域进行指数滤波,得到滤波后的新能量谱掩膜矩阵θ;
矩阵重建单元,与所述控制器连接,用于将所述新能量谱掩膜矩阵θ与所述平移谱矩阵
Figure PCTCN2022129187-appb-000019
进行点乘运算,得到重建谱矩阵E;
第二矩阵处理单元,与所述控制器连接,用于将所述重建谱矩阵E依次进行频谱平移和逆傅里叶变换,得到重建后的灰度图像Ψ;
非线性拉伸单元,与所述控制器连接,用于对所述重建后的灰度图像Ψ进行非线性拉伸变换,得到最终的处理图像。
根据本发明实施例的第三方面,提供一种电子设备,其中包括:
存储器,所述存储器内存储有所述的摩尔纹祛除方法的程序或指令;
处理器,所述处理器与所述存储器耦合,用于执行所述存储器中的程序或指令,以使所述电子设备执行所述的摩尔纹祛除方法。
本发明提供的摩尔纹祛除方法,通过对文字图像的能量谱分析,标记能量二值图像的连通区域,并且对非最大的连通区域进行指数滤波,计算能量谱的掩膜表示,最终通过谱矩阵的重建将摩尔纹从文字图像上削弱甚至完全祛除。该方法不依赖于训练数据集,鲁棒性良好;相比于基于深度学习网络的算法,计算量大大减少,特别适合手机等嵌入式设备。
附图说明
图1为本发明实施例提供的摩尔纹祛除方法的流程示意图;
图2为本发明实施例中,原始谱矩阵Y的示意图;
图3为本发明实施例中,滤波后的新能量谱掩膜矩阵θ的示意图;
图4为本发明实施例中,处理前文字图像的示意图;
图5为本发明实施例中,处理后文字图像的示意图;
图6为本发明实施例提供的摩尔纹祛除装置的结构示意图;
图7为本发明实施例提供的电子设备的结构示意图。
具体实施方式
下面结合附图和具体实施例对本发明的技术内容进行详细具体的说明。
<第一实施例>
请参照图1所示,为本发明实施例提供的一种文字图像的摩尔纹祛除方法,至少包括以下步骤:
S1:将文字图像转化为灰度图像,并对灰度图像进行归一化处理,得到灰度图像矩阵D。
具体的,在本实施例中,当需要祛除文字图像的摩尔纹时,需要对文字图像进行预处理,该预处理的步骤为:首先将文字图像转化为灰度图像,然后,对灰度图像进行归一化处理。其中,归一化的目的是避免后续运算中的数据溢出,而归一化处理的过程为常规的技术手段,在此不再赘述。
S2:将灰度图像矩阵D依次进行二维傅里叶变换、频谱平移和取模操作,得到平移谱矩阵
Figure PCTCN2022129187-appb-000020
具体的,包括步骤S21~S22:
S21:将所述灰度图像矩阵D进行二维傅里叶变换,得到原始谱矩阵Y。
具体的,参照图2所示,在本实施例中,设归一化后的灰度图像矩阵 D为m×n的二维矩阵,将灰度图像矩阵D按照下列公式进行二维傅里叶变换,以得到原始谱矩阵Y:
Figure PCTCN2022129187-appb-000021
其中,
Figure PCTCN2022129187-appb-000022
i为虚数单位,d (j,k)为矩阵D中j行k列的元素,y (p,q)为原始谱矩阵Y的p行q列的元素。
当得到原始谱矩阵Y后,以谱矩阵的行和列的二分之一的交点作为原点O划分出第一象限、第二象限、第三象限和第四象限,即图2中一、二、三、四所示的位置。可以理解的是,在原始谱矩阵Y中,位于原点O附近的区域为高频区域,而位于原始谱矩阵Y四角的位置为低频区域。
S22:将所述原始谱矩阵Y进行频谱平移,得到平移后的平移谱矩阵
Figure PCTCN2022129187-appb-000023
具体的,将原始谱矩阵Y的第一象限与第三象限交换,并将原始谱矩阵Y的第二象限与第四象限交换得到平移后的平移谱矩阵
Figure PCTCN2022129187-appb-000024
S3:对所述平移谱矩阵
Figure PCTCN2022129187-appb-000025
中所有元素取模,然后进行二值化处理,得到能量谱掩膜矩阵Z。
具体的,包括步骤S31~S32:
S31:取出所述平移谱矩阵
Figure PCTCN2022129187-appb-000026
中各个元素的模,组成同维度的能量谱矩阵X。由此,从同维度的能量谱矩阵X中,可以大致看出哪些区域的能量较高,哪些区域的能量较低。
S32:当建立能量谱矩阵X后,通过下列公式对能量谱矩阵X做阈值判决,得到二值化的能量谱掩膜矩阵Z,即:
Figure PCTCN2022129187-appb-000027
其中,z (i,j)和x(i,j)分别是能量谱掩膜矩阵Z和能量谱矩阵X在i行j列的元素,T为判决阈值,通过对图像亮度进行整体平衡得到,例如:通过亮度累积直方图或者用otsu法获取。
由此,通过二值化处理后,能够确认能量谱掩膜矩阵Z中的能量突出区域,其中,所述能量突出区域包括因摩尔纹的高频纹路而导致的第一能量突出区域和因文字固有纹路而导致的第二能量突出区域。
S4:对能量谱掩膜矩阵Z进行形态学闭运算,并对连通区域进行标记。
具体的,包括步骤S41~S42:
S41:使用盘式结构对能量谱掩膜矩阵Z进行形态学闭运算,其中,盘式结构具有设定半径,本实施例中,盘式结构的半径推荐为8,当然也可以是其他数值。由此,通过对能量谱掩膜矩阵Z进行形态学闭运算,从而可以去除能量突出区域中的“小型能量孤岛”,进而减少“小型能量孤岛”对后续摩尔纹高频纹路导致的第一能量突出区域的影响。其中,该“小型能量孤岛”指的是:具有一定的高能量点,但区域较小且较为分散的区域。
S42:根据形态学闭运算,通过范洪填充算法对能量谱掩膜矩阵Z的连通区域进行标记。其中,范洪填充算法为目前已知的常规算法,在此不再赘述。由此,通过对能量谱掩膜矩阵Z的连通区域进行标记,从而可以清晰地区分第一能量突出区域与第二能量突出区域的位置,以便于后续对第一能量突出区域进行能量削弱。
S5:获取各连通区域的所包含元素的坐标以及所有连通区域的数量,对所有非最大连通区域进行指数滤波,得到滤波后的新能量谱掩膜矩阵θ。
可以理解的是,由于最大连通区域为整个图像前景的最主要部分,如果对其进行处理后,则可能导致整个图像的亮度、色差出现较大的变化。因此,为避免对图像造成太大的影响,对最大的连通区域不做处理,仅对所有的非最大连通区域进行指数滤波。
具体的,包括步骤S51~S53:
S51:估算任意一个非最大连通区域的中心坐标。
具体的,通过聚类算法对如下代价函数进行迭代,以估算出任意一个非最大连通区域S τ的中心坐标C τ(x o,y o)。
Figure PCTCN2022129187-appb-000028
其中,
Figure PCTCN2022129187-appb-000029
为连通区域元素集合S τ中的第j个元素,其坐标为(x j,y j);C τ(x o,y o)为该连通区域的中心坐标;非最大连通区域S τ内共包含H个元素;‖*‖为距离度量的运算符,包括但不限于欧氏距离、马氏距离等。
S52:遍历非最大连通区域中的所有的元素坐标,根据非最大连通区域中各元素坐标与中心坐标之间的距离进行指数滤波,以完成一个非最大连通区域的指数滤波;
具体的,在本实施例中,遍历该非最大连通区域S τ中的所有的元素坐标,j=0,...,H-1,按照下式进行指数滤波;
Figure PCTCN2022129187-appb-000030
其中,α=range(x∈S τ),β=range(y∈S τ),x∈S τ和y∈S τ分别表示连通区域元素集合S τ中所有元素的x坐标集合和y坐标集合,函数range(w)=max(w)-min(w)。
可以理解的是,当经过指数滤波后,在该非最大连通区域内,不再是0或1的分布情况,而是0~1之间的数值,例如:0.3、0.5、0.8、0.9等不同的数值,从而使得在非最大连通区域内,各个坐标点所代表的是对特定频率范围上原始能量不同的降低,甚至消除。由此,通过指数滤波的方式可以削弱或消除第一能量突出区域的能量参数,从而达到削弱或消除摩尔纹的效果。
S53:遍历所有非最大连通区域,以通过相同的方式完成所有非最大连通区域的指数滤波,得到滤波后的新能量谱掩膜矩阵θ。
具体的,重复步骤S51~S52,直至对所有非最大连通区域完成指数滤波,从而得到滤波后的新能量谱掩膜矩阵θ。参照图3所示,在滤波后的新能量谱掩膜矩阵θ中。同时,图3中的数值仅作为示例,不构成对新能量谱掩膜矩阵θ的数据限定。
S6:将新能量谱掩膜矩阵θ与平移谱矩阵
Figure PCTCN2022129187-appb-000031
进行点乘运算,得到重建谱矩阵E;
具体的,可表示为
Figure PCTCN2022129187-appb-000032
其中,⊙为点乘运算,即两个矩阵元素分别相乘。由此,可以消除文字图像的摩尔纹高频纹路干扰。
S7:将重建谱矩阵E依次进行频谱平移和逆傅里叶变换,得到重建后的灰度图像Ψ;
具体的包括步骤S71~S72:
S71:将重建谱矩阵E进行频谱平移。
具体的,将重建谱矩阵E的第一象限与第三象限交换,并将重建谱矩阵E的第二象限与第四象限交换,得到平移后的谱矩阵
Figure PCTCN2022129187-appb-000033
S72:通过下列公式对平移后的谱矩阵
Figure PCTCN2022129187-appb-000034
进行二维逆傅里叶变换,得到重建后的灰度图像Ψ;
Figure PCTCN2022129187-appb-000035
其中,
Figure PCTCN2022129187-appb-000036
i为虚数单位,
Figure PCTCN2022129187-appb-000037
为矩阵
Figure PCTCN2022129187-appb-000038
中j行k列的元素,ψ (p,q)为灰度图像Ψ的p行q列的元素。
由此,通过将重建谱矩阵E依次进行频谱平移和逆傅里叶变换后,能够将处理后的图像重新变换回灰度图像。
S8:对重建后的灰度图像Ψ进行非线性拉伸变换,得到最终的处理图像。
具体的,由于对原灰度图像经过步骤S1~S7处理后,重建后的灰度图像Ψ可能会存在亮度不均匀的情况,因此,通过下列公式对重建后的灰度图像Ψ进行迭代,以完成对重建后的灰度图像Ψ的非线性拉伸变换;
Ψ i+1=Ψ i+(1-Ψ i)⊙Ψ i*∈
其中,Ψ i+1为第i次迭代的输出灰度图像,Ψ i第i次迭代的输入灰度图像,∈为调节灵敏度参数。
由此,可以调节亮度的均衡,进而提高文字在背景中的对比度,图4位处理前文字图像的示意图,图5为经过步骤S1~S8处理后的文字图像的示意图,对比图4和图5可以看出,文字图像的摩尔纹被减弱,且文字部分得以较为完整的保留,没有对文字信息造成明显的破坏。
综上所述,本发明实施例提供的摩尔纹祛除方法,通过对文字图像的能量谱分析,标记能量二值图像的连通区域,并且对非最大的连通区域进行指数滤波,计算能量谱的掩膜表示,最终通过谱矩阵的重建将摩尔纹从文字图像上削弱甚至完全祛除。该方法不依赖于训练数据集,鲁棒性良好;相比于基于深度学习网络的算法,计算量大大减少,适合手机等嵌入式设备。
<第二实施例>
如图6所示,本发明实施例还提供一种摩尔纹祛除装置,至少包括: 图像转化单元1、第一矩阵处理单元2、二值化处理单元3、连通区域标记单元4、指数滤波单元5、矩阵重建单元6、第二矩阵处理单元7、非线性拉伸单元8以及控制器9。
具体的,图像转化单元1与控制器9连接,用于将文字图像转化为灰度图像,并对灰度图像进行归一化处理,得到灰度图像矩阵D;第一矩阵处理单元2与控制器9连接,用于对灰度图像矩阵D依次进行二维傅里叶变换、频谱平移和取模运算,得到平移后的谱矩阵
Figure PCTCN2022129187-appb-000039
二值化处理单元3与控制器9连接,用于对平移后的谱矩阵
Figure PCTCN2022129187-appb-000040
进行二值化处理,得到能量谱掩膜矩阵Z;连通区域标记单元4与控制器9连接,用于对能量谱掩膜矩阵Z进行形态学闭运算,并对连通区域进行标记;指数滤波单元5与控制器9连接,用于获取各连通区域所包含元素的坐标以及所有连通区域的元素数量,对所有非最大连通区域进行指数滤波,得到滤波后的新能量谱掩膜矩阵θ;矩阵重建单元6与控制器9连接,用于将新能量谱掩膜矩阵θ与平移谱矩阵
Figure PCTCN2022129187-appb-000041
进行点乘运算,得到重建谱矩阵E;第二矩阵处理单元7与控制器9连接,用于将重建谱矩阵E依次进行频谱平移和逆傅里叶变换,得到重建后的灰度图像Ψ;非线性拉伸单元8与控制器9连接,用于对重建后的灰度图像Ψ进行非线性拉伸变换,得到最终的处理图像。
<第三实施例>
本发明实施例还提供一种电子设备,至少包括存储器和处理器。其中,存储器内存储有的摩尔纹祛除方法的程序或指令;处理器与存储器耦合,用于执行存储器中的程序或指令,以使电子设备执行上述第一实施例中的摩尔纹祛除方法。这里所说的电子设备是指可以在移动环境中使用,支持GSM、EDGE、TD_SCDMA、TDD_LTE、FDD_LTE等多种通信制式的计算机设备,包括智能手机、笔记本电脑、平板电脑、车载电脑等。
如图7所示,该电子设备至少包括处理器和存储器,还可以根据实际需要进一步包括通信组件、传感器组件、电源组件、多媒体组件及输入/输出接口。其中,存储器、通信组件、传感器组件、电源组件、多媒体组件及输入/输出接口均与该处理器连接。存储器可以是静态随机存取存储器(SRAM)、电可擦除可编程只读存储器(EEPROM)、可擦除可编程只读存储器(EPROM)、可编程只读存储器(PROM)、只读存储器(ROM)、磁存储器、快闪存储器等,处理器可以是中央处理器(CPU)、 图形处理器(GPU)、现场可编程逻辑门阵列(FPGA)、专用集成电路(ASIC)、数字信号处理(DSP)芯片等。其它通信组件、传感器组件、电源组件、多媒体组件等均可以采用通用部件实现,在此就不具体说明了。
上面对本发明所提供的文字图像的摩尔纹祛除方法、装置及电子设备进行了详细的说明。对本领域的一般技术人员而言,在不背离本发明实质内容的前提下对它所做的任何显而易见的改动,都将构成对本发明专利权的侵犯,将承担相应的法律责任。

Claims (10)

  1. 一种文字图像的摩尔纹祛除方法,其特征在于包括以下步骤:
    将文字图像转化为灰度图像,并对所述灰度图像进行归一化处理,得到灰度图像矩阵D;
    将所述灰度图像矩阵D依次进行二维傅里叶变换、频谱平移和取模运算,得到平移后的谱矩阵
    Figure PCTCN2022129187-appb-100001
    对所述平移后的谱矩阵
    Figure PCTCN2022129187-appb-100002
    进行二值化处理,得到能量谱掩膜矩阵Z,以确认能量谱掩膜矩阵Z中的能量突出区域,所述能量突出区域包括因摩尔纹的高频纹路而导致的第一能量突出区域和因文字固有纹路而导致的第二能量突出区域;
    对所述能量谱掩膜矩阵Z进行形态学闭运算,并对连通区域进行标记,以对所述能量突出区域进行能量优化,并区分所述第一能量突出区域与所述第二能量突出区域的位置;
    获取各所述连通区域所包含元素的坐标以及所有连通区域的元素数量,对所有非最大连通区域进行指数滤波,得到滤波后的新能量谱掩膜矩阵θ,以削弱或消除所述第一能量突出区域的能量参数;
    将所述新能量谱掩膜矩阵θ与所述平移谱矩阵
    Figure PCTCN2022129187-appb-100003
    进行点乘运算,得到重建谱矩阵E,以消除文字图像的摩尔纹高频纹路干扰;
    将所述重建谱矩阵E依次进行频谱平移和逆傅里叶变换,得到重建后的灰度图像Ψ。
  2. 如权利要求1所述的摩尔纹祛除方法,其特征在于,所述对所述平移谱矩阵
    Figure PCTCN2022129187-appb-100004
    进行二值化处理,得到能量谱掩膜矩阵Z,具体包括:
    取出所述平移谱矩阵
    Figure PCTCN2022129187-appb-100005
    中各个元素的模,组成同维度的能量谱矩阵X;
    通过下列公式对所述能量谱矩阵X做阈值判决,得到二值化的能量谱掩膜矩阵Z,即:
    Figure PCTCN2022129187-appb-100006
    其中,z (i,j)和x(i,j)分别是能量谱掩膜矩阵Z和能量谱矩阵X在i行j列的元素,T为判决阈值,通过对图像亮度进行整体平衡得到。
  3. 如权利要求1所述的摩尔纹祛除方法,其特征在于,所述获取各 所述连通区域所包含元素的坐标以及所有连通区域的元素数量,对所有非最大连通区域进行指数滤波,得到滤波后的新能量谱掩膜矩阵θ,具体包括:
    估算任意一个非最大连通区域的中心坐标;
    遍历所述非最大连通区域中的所有的元素坐标,根据所述非最大连通区域中各元素坐标与所述中心坐标之间的距离进行指数滤波,以完成一个非最大连通区域的指数滤波;
    遍历所有非最大连通区域,以通过相同的方式完成所有非最大连通区域的指数滤波,得到滤波后的新能量谱掩膜矩阵θ。
  4. 如权利要求3所述的摩尔纹祛除方法,其特征在于,所述估算任意一个非最大连通区域的中心坐标,具体包括:
    通过聚类算法对如下代价函数进行迭代,以估算出任意一个非最大连通区域S τ的中心坐标C τ(x o,y o);
    Figure PCTCN2022129187-appb-100007
    其中,
    Figure PCTCN2022129187-appb-100008
    为连通区域元素集合S τ中的第j个元素,其坐标为(x j,y j);C τ(x o,y o)为该连通区域的中心坐标;所述非最大连通区域S τ内共包含H个元素;‖*‖为距离度量的运算符,至少包括欧氏距离或马氏距离运算。
  5. 如权利要求4所述的摩尔纹祛除方法,其特征在于,所述遍历所述非最大连通区域中的所有的元素坐标,根据所述非最大连通区域中各元素坐标与所述中心坐标之间的距离进行指数滤波,以完成一个非最大连通区域的指数滤波,具体包括:
    遍历所述非最大连通区域S τ中的所有的元素坐标,j=0,...,H-1,按照下式进行指数滤波;
    Figure PCTCN2022129187-appb-100009
    其中,α=range(x∈S τ),β=range(y∈S τ),x∈S τ和y∈S τ分别表示连通区域元素集合S τ中所有元素的x坐标集合和y坐标集合,函数range(w)=max(w)-min(w)。
  6. 如权利要求1所述的摩尔纹祛除方法,其特征在于,所述将所述重建谱矩阵E依次进行频谱平移和逆傅里叶变换,得到重建后的灰度图像Ψ,具体包括:
    将所述重建谱矩阵E的第一象限与第三象限交换,并将所述重建谱矩阵E的第二象限与第四象限交换,得到平移后的谱矩阵
    Figure PCTCN2022129187-appb-100010
    通过下列公式对所述平移后的谱矩阵
    Figure PCTCN2022129187-appb-100011
    进行二维逆傅里叶变换,得到重建后的灰度图像Ψ;
    Figure PCTCN2022129187-appb-100012
    其中,
    Figure PCTCN2022129187-appb-100013
    i为虚数单位,
    Figure PCTCN2022129187-appb-100014
    为矩阵
    Figure PCTCN2022129187-appb-100015
    中j行k列的元素,ψ (p,q)为灰度图像Ψ的p行q列的元素。
  7. 如权利要求1所述的摩尔纹祛除方法,其特征在于还包括以下步骤:
    对所述重建后的灰度图像Ψ进行非线性拉伸变换,得到最终的处理图像。
  8. 如权利要求7所述的摩尔纹祛除方法,其特征在于,所述对所述重建后的灰度图像Ψ进行非线性拉伸变换,得到最终的处理图像,具体包括:
    通过下列公式对所述重建后的灰度图像Ψ进行迭代,以完成对所述重建后的灰度图像Ψ的非线性拉伸变换;
    Ψ i+1=Ψ i+(1-Ψ i)⊙Ψ i*∈
    其中,Ψ i+1为第i次迭代的输出灰度图像,Ψ i第i次迭代的输入灰度图像,∈为调节灵敏度参数。
  9. 一种摩尔纹祛除装置,包括控制器,其特征在于还包括:
    图像转化单元,与控制器连接,用于将文字图像转化为灰度图像,并对所述灰度图像进行归一化处理,得到灰度图像矩阵D;
    第一矩阵处理单元,与所述控制器连接,用于对所述灰度图像矩阵D依次进行二维傅里叶变换、频谱平移和取模运算,得到平移后的谱矩阵
    Figure PCTCN2022129187-appb-100016
    二值化处理单元,与所述控制器连接,用于对所述平移后的谱矩阵
    Figure PCTCN2022129187-appb-100017
    进行二值化处理,得到能量谱掩膜矩阵Z;
    连通区域标记单元,与所述控制器连接,用于对所述能量谱掩膜矩阵Z进行形态学闭运算,并对连通区域进行标记;
    指数滤波单元,与所述控制器连接,用于获取各所述连通区域所包含元素的坐标以及所有连通区域的元素数量,对所有非最大连通区域进行指数滤波,得到滤波后的新能量谱掩膜矩阵θ;
    矩阵重建单元,与所述控制器连接,用于将所述新能量谱掩膜矩阵θ与所述平移谱矩阵
    Figure PCTCN2022129187-appb-100018
    进行点乘运算,得到重建谱矩阵E;
    第二矩阵处理单元,与所述控制器连接,用于将所述重建谱矩阵E依次进行频谱平移和逆傅里叶变换,得到重建后的灰度图像Ψ;
    非线性拉伸单元,与所述控制器连接,用于对所述重建后的灰度图像Ψ进行非线性拉伸变换,得到最终的处理图像。
  10. 一种电子设备,其特征在于包括:
    存储器,所述存储器内存储有权利要求1所述的摩尔纹祛除方法的程序或指令;
    处理器,所述处理器与所述存储器耦合,用于执行所述存储器中的程序或指令,以使所述电子设备执行权利要求1所述的摩尔纹祛除方法。
PCT/CN2022/129187 2021-11-05 2022-11-02 一种文字图像的摩尔纹祛除方法、装置及电子设备 WO2023078285A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111308232.X 2021-11-05
CN202111308232.XA CN114022383A (zh) 2021-11-05 2021-11-05 一种文字图像的摩尔纹祛除方法、装置及电子设备

Publications (1)

Publication Number Publication Date
WO2023078285A1 true WO2023078285A1 (zh) 2023-05-11

Family

ID=80061806

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/129187 WO2023078285A1 (zh) 2021-11-05 2022-11-02 一种文字图像的摩尔纹祛除方法、装置及电子设备

Country Status (2)

Country Link
CN (1) CN114022383A (zh)
WO (1) WO2023078285A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117078233A (zh) * 2023-10-17 2023-11-17 深圳市城市交通规划设计研究中心股份有限公司 一种基于路网养护综合评价指标的养护决策方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114022383A (zh) * 2021-11-05 2022-02-08 瞬联软件科技(北京)有限公司 一种文字图像的摩尔纹祛除方法、装置及电子设备
CN115131234B (zh) * 2022-06-15 2023-09-19 西北大学 一种基于两阶段神经网络的数字壁画修复方法
CN117372283A (zh) * 2023-11-06 2024-01-09 上海衡亮电子科技股份有限公司 一种去除摩尔纹的方法与装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2014174803A (ja) * 2013-03-11 2014-09-22 Omron Corp パターン認識のためのモアレ除去方法およびこの方法を用いたモアレ除去装置ならびにプログラム
CN108389164A (zh) * 2018-01-31 2018-08-10 深圳市商巨视觉技术有限公司 基于频域分析的去除图像摩尔纹的方法
CN111383189A (zh) * 2018-12-29 2020-07-07 合肥欣奕华智能机器有限公司 一种去除摩尔纹的方法及装置、图像显示器
CN111383186A (zh) * 2018-12-29 2020-07-07 Tcl集团股份有限公司 一种图像处理方法、装置及终端设备
CN114022383A (zh) * 2021-11-05 2022-02-08 瞬联软件科技(北京)有限公司 一种文字图像的摩尔纹祛除方法、装置及电子设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2014174803A (ja) * 2013-03-11 2014-09-22 Omron Corp パターン認識のためのモアレ除去方法およびこの方法を用いたモアレ除去装置ならびにプログラム
CN108389164A (zh) * 2018-01-31 2018-08-10 深圳市商巨视觉技术有限公司 基于频域分析的去除图像摩尔纹的方法
CN111383189A (zh) * 2018-12-29 2020-07-07 合肥欣奕华智能机器有限公司 一种去除摩尔纹的方法及装置、图像显示器
CN111383186A (zh) * 2018-12-29 2020-07-07 Tcl集团股份有限公司 一种图像处理方法、装置及终端设备
CN114022383A (zh) * 2021-11-05 2022-02-08 瞬联软件科技(北京)有限公司 一种文字图像的摩尔纹祛除方法、装置及电子设备

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117078233A (zh) * 2023-10-17 2023-11-17 深圳市城市交通规划设计研究中心股份有限公司 一种基于路网养护综合评价指标的养护决策方法
CN117078233B (zh) * 2023-10-17 2024-03-26 深圳市城市交通规划设计研究中心股份有限公司 一种基于路网养护综合评价指标的养护决策方法

Also Published As

Publication number Publication date
CN114022383A (zh) 2022-02-08

Similar Documents

Publication Publication Date Title
WO2023078285A1 (zh) 一种文字图像的摩尔纹祛除方法、装置及电子设备
CN107561738B (zh) 基于fcn的tft-lcd表面缺陷快速检测方法
WO2022170706A1 (zh) 用于模具监视的缺陷检测方法、装置、设备及介质
CN108805023B (zh) 一种图像检测方法、装置、计算机设备及存储介质
CN110930390B (zh) 基于半监督深度学习的芯片管脚缺失检测方法
CN108154149B (zh) 基于深度学习网络共享的车牌识别方法
CN107169979A (zh) 一种改进Canny算子的图像边缘检测方法
CN113362320A (zh) 基于深度注意力网络的晶圆表面缺陷模式检测方法
CN112132196B (zh) 一种结合深度学习和图像处理的烟盒缺陷识别方法
CN110263809B (zh) 池化特征图处理方法、目标检测方法、系统、装置和介质
CN113781406B (zh) 电子元器件的划痕检测方法、装置及计算机设备
US11348349B2 (en) Training data increment method, electronic apparatus and computer-readable medium
CN107506795A (zh) 一种面向图像匹配的局部灰度直方图特征描述子建立方法和图像匹配方法
JP2021531571A (ja) 証明書画像抽出方法及び端末機器
CN116342610B (zh) 一种钢结构装配式建筑焊接异常检测方法
CN111680690A (zh) 一种文字识别方法及装置
US20130064468A1 (en) Methods and Apparatus for Image Analysis and Modification Using Fast Sliding Parabola Erosian
WO2021189770A1 (zh) 基于人工智能的图像增强处理方法、装置、设备及介质
Li et al. Electronic product surface defect detection based on a MSSD network
CN110827209A (zh) 一种联合颜色与深度信息的自适应深度图像修复方法
US11270152B2 (en) Method and apparatus for image detection, patterning control method
Stanciu et al. Contrast enhancement influences the detection of gradient based local invariant features and the matching of their descriptors
US20220122244A1 (en) Defect image generation method for deep learning and system therefor
CN112991236B (zh) 一种基于模板的图像增强方法及装置
CN107680068A (zh) 一种考虑图像自然度的数字图像增强方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22889307

Country of ref document: EP

Kind code of ref document: A1