WO2020077866A1

WO2020077866A1 - Moire-based image recognition method and apparatus, and device and storage medium

Info

Publication number: WO2020077866A1
Application number: PCT/CN2018/124583
Authority: WO
Inventors: 陈粉玉; 韩冰
Original assignee: 平安科技（深圳）有限公司
Priority date: 2018-10-17
Filing date: 2018-12-28
Publication date: 2020-04-23
Also published as: CN109558794A

Abstract

The invention relates to the technical field of image recognition. Disclosed are a moire-based image recognition method and apparatus, and a computer device and a storage medium. The moire-based image recognition method comprises: acquiring multiple images to be processed, and packaging them into an image set, wherein the image set comprises an image with moire and an image without moire; extracting moire features included in any one image in the image set to form a training sample; taking the training sample as an input parameter, and performing SVM training in an SVM model to obtain an image recognition model; and performing moire recognition on an unknown image according to the image recognition model. With regard to the problem of conducting fraud using a reproduced picture in the process of a financial transaction, moire recognition is performed using an LBP algorithm and training is performed by means of an SVM model, thus improving the speed and accuracy of recognition of a reproduced photograph.

Description

基于摩尔纹的图像识别方法、装置、设备和存储介质Image recognition method, device, equipment and storage medium based on moiré

本申请要求于2018年10月17日提交中国专利局、申请号为201811206940.0、发明名称为“基于摩尔纹的图像识别方法、装置、设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。This application requires the priority of the Chinese patent application submitted to the Chinese Patent Office on October 17, 2018, with the application number 201811206940.0 and the invention titled "Image recognition method, device, equipment and storage medium based on moiré", all of its content Incorporated by reference in this application.

技术领域Technical field

本申请涉及图像识别技术领域，尤其涉及一种基于摩尔纹的图像识别方法、装置、设备和存储介质。The present application relates to the field of image recognition technology, and in particular, to an image recognition method, device, device, and storage medium based on moiré.

背景技术Background technique

在金融活动过程中，往往需要对使用者进行人脸识别以确定身份。然而在进行人脸识别过程中，某些人使用翻拍的人脸图像来冒充真实的人脸以实现欺诈的目的。目前在对翻拍照片识别过程中主要采用通过随机发出一些随机性的指令动作，例如眨眼、左右摇头以及上下摇头之类的进行识别；或者采用红外线对于人体温度进行识别。In the process of financial activities, it is often necessary to perform face recognition on users to determine their identities. However, in the process of face recognition, some people use the remake of the face image to impersonate the real face to achieve the purpose of fraud. At present, in the process of recognizing remakes, the main use is to randomly issue some random command actions, such as blinking, shaking his head left and right, and shaking his head up and down; or using infrared to identify the temperature of the human body.

但是，在使用随机指令动作进行识别时存在着响应时差，导致需要使用者多次重复动作才能完成识别，而采用红外线人体温度识别则会收到周围环境的干扰降低识别的准确率。However, there is a time difference in response when using random command actions to recognize, which requires the user to repeat the action many times to complete the recognition, and using infrared human body temperature recognition will receive interference from the surrounding environment and reduce the accuracy of recognition.

发明内容Summary of the invention

基于此，有必要针对现有翻拍照片识别度低，不能及时有效对使用者身份进行识别的问题，提供一种基于摩尔纹的图像识别方法、装置、设备和存储介质。Based on this, it is necessary to provide an image recognition method, device, equipment, and storage medium based on Moiré for the problem that the existing remakes have low recognition and cannot effectively identify the user's identity in time.

一种基于摩尔纹的图像识别方法，包括：An image recognition method based on moiré includes:

获取待处理的数个图像，打包成一图像集，所述图像集中包括带有摩尔纹的图像和不带有摩尔纹的图像；Obtain several images to be processed and pack them into an image set, where the image set includes images with moiré and images without moiré;

提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本；Extracting moiré features contained in any image in the image set to form a training sample;

将所述训练样本作为入参，在支持向量机SVM模型中进行SVM训练，得到图像识别模型；Using the training samples as input parameters, performing SVM training in the support vector machine SVM model to obtain an image recognition model;

接收上传的未知图像，根据所述图像识别模型对所述未知图像进行摩尔纹识别，以确定所述图像是否为翻拍图像。Receiving the uploaded unknown image, and performing moiré recognition on the unknown image according to the image recognition model to determine whether the image is a remake image.

一种基于摩尔纹的图像识别装置，包括如下模块：An image recognition device based on moiré includes the following modules:

图像集形成模块，设置为获取待处理的数个图像，打包成一图像集，所述图像集中包括带有摩尔纹的图像和不带有摩尔纹的图像；The image set forming module is set to obtain several images to be processed and packaged into an image set, the image set includes images with moiré and images without moiré;

训练样本生成模块，设置为提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本；The training sample generation module is set to extract the moiré features contained in any image in the image set to form a training sample;

识别模型生成模块，设置为将所述训练样本作为入参，在支持向量机SVM模型中进行SVM训练，得到图像识别模型；The recognition model generation module is set to use the training samples as input parameters and perform SVM training in the support vector machine SVM model to obtain an image recognition model;

未知图像识别模块，设置为接收上传的未知图像，根据所述图像识别模型对所述未知图像进行摩尔纹识别，以确定所述图像是否为翻拍图像。The unknown image recognition module is configured to receive the uploaded unknown image and perform moiré recognition on the unknown image according to the image recognition model to determine whether the image is a remake image.

一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行上述基于摩尔纹的图像识别方法的步骤。A computer device includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor causes the processor to perform the above-mentioned Moiré-based image recognition method A step of.

一种存储有计算机可读指令的存储介质，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行上述基于摩尔纹的图像识别方法的步骤。A storage medium storing computer-readable instructions, which when executed by one or more processors, causes the one or more processors to perform the steps of the above-mentioned moiré-based image recognition method.

上述基于摩尔纹的图像识别方法、装置、计算机设备和存储介质，包括获取待处理的数个图像，打包成一图像集，所述图像集中包括带有摩尔纹的图像和不带有摩尔纹的图像；提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本；将所述训练样本作为入参，在支持向量机SVM模型中进行SVM训练，得到图像识别模型；接收上传的未知图像，根据所述图像识别模型对所述未知图像进行摩尔纹识别，以确定所述图像是否为翻拍图像。本技术方案针对在金融交易过程中存在使用翻拍图片进行欺诈的问题，采用LBP算法对摩尔纹进行识别，并经过SVM模型训练，提升对翻拍照片识别的速度和准确度。The above-mentioned moiré-based image recognition method, device, computer equipment, and storage medium include acquiring several images to be processed and packaging them into an image set, which includes images with moiré and images without moiré ; Extract the moiré features contained in any image in the image set to form a training sample; use the training sample as an input parameter, and perform SVM training in the support vector machine SVM model to obtain an image recognition model; receive the uploaded unknown Image, performing moiré recognition on the unknown image according to the image recognition model to determine whether the image is a remake image. This technical solution aims at the problem of fraudulent use of remake pictures in the process of financial transactions. The LBP algorithm is used to identify moiré patterns, and the SVM model is trained to improve the speed and accuracy of reprint picture recognition.

附图说明BRIEF DESCRIPTION

通过阅读下文优选实施方式的详细描述，各种其他的优点和益处对于本领域普通技术人员将变得清楚明了。附图仅用于示出优选实施方式的目的，而并不认为是对本申请的限制。By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments, and are not considered to limit the present application.

图1为本申请一种基于摩尔纹的图像识别方法的整体流程图；FIG. 1 is an overall flowchart of an image recognition method based on moiré of this application;

图2为本申请一种基于摩尔纹的图像识别方法中的图像集形成过程示意图；2 is a schematic diagram of an image set forming process in an image recognition method based on moiré of this application;

图3为本申请一种基于摩尔纹的图像识别方法中的摩识别样本生成过程示意图；FIG. 3 is a schematic diagram of a process of generating a friction recognition sample in a moiré-based image recognition method of this application;

图4为本申请一种基于摩尔纹的图像识别装置的结构图。FIG. 4 is a structural diagram of an image recognition device based on moiré of the present application.

具体实施方式detailed description

为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。In order to make the purpose, technical solutions and advantages of the present application more clear, the following describes the present application in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

本技术领域技术人员可以理解，除非特意声明，这里使用的单数形式“一”、“一个”、“所述”和“该”也可包括复数形式。应该进一步理解的是，本申请的说明书中使用的措辞“包括”是指存在所述特征、整数、步骤、操作、元件和/或组件，但是并不排除存在或添加一个或多个其他特征、整数、步骤、操作、元件、组件和/或它们的组。Those skilled in the art can understand that unless specifically stated, the singular forms "a", "an", "said" and "the" used herein may also include the plural forms. It should be further understood that the word "comprising" used in the specification of this application refers to the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, Integers, steps, operations, elements, components, and / or their groups.

图1为本申请一个实施例中的基于摩尔纹的图像识别方法的流程图，如图1所示，一种基于摩尔纹的图像识别方法，包括：FIG. 1 is a flowchart of a moiré-based image recognition method in an embodiment of the present application. As shown in FIG. 1, a moiré-based image recognition method includes:

S1，获取待处理的数个图像，打包成一图像集，所述图像集中包括带有摩尔纹的图像和不带有摩尔纹的图像；S1. Acquire several images to be processed and pack them into an image set, where the image set includes images with moiré and images without moiré;

具体的，调用设置在检测终端的图像采集器对两组图像分别进行图像采集，其中第一组图像为真实人脸，第二组图像为翻拍照片，图像采集器可以是CCD图像采集器，其主要由以下几个部分组成：前端光学装置，CCD图像采集模块，模数转换模块，FPGA预处理模块，Flash程序存储模块，DSP图像处理模块，SDRAM数据存储模块，图像显示模块，和图像处理器。Specifically, the image collector set at the detection terminal is called to separately collect images of two groups of images, wherein the first group of images is a real face, the second group of images is a remake photo, and the image collector may be a CCD image collection It is mainly composed of the following parts: front-end optical device, CCD image acquisition module, analog-to-digital conversion module, FPGA preprocessing module, Flash program storage module, DSP image processing module, SDRAM data storage module, image display module, and Image processor.

在进行图像采集时，在CCD图像采集器上设置感光元件对环境光强进行感应，在后端PC机的处理器中设置有光强阈值，通常情况下光强阈值设置为200lx，当环境光强小于200lx时启动闪光灯，大于200lx时则不用闪光灯。During image acquisition, the photosensitive element is set on the CCD image collector to sense the ambient light intensity, and the light intensity threshold is set in the processor of the back-end PC. Normally, the light intensity threshold is set to 200 lx, when the ambient light The flash is activated when the intensity is less than 200lx, and the flash is not used when it is greater than 200lx.

S2，提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本；S2, extracting moiré features contained in any image in the image set to form a training sample;

具体的，在进行摩尔纹特征提取的时采用LBP算法进行提取，局部二值模式(英文：Local binary patterns，缩写：LBP)是机器视觉领域中用于分类的一种特征，于1994年被提出。局部二值模式在纹理分类问题上是一个非常强大的特征；如果局部二值模式特征与方向梯度直方图结合，则可以十分有效的提升检测效果。局部二值模式是一个简单但非常有效的纹理运算符。LBP最重要的属性是对诸如光照变化等造成的灰度变化的鲁棒性，它的另外一个重要特性是它的计算简单，这使得它可以对图像进行实时分析。Specifically, when extracting moiré features, the LBP algorithm is used for extraction. The local binary pattern (English: Local binary patterns, abbreviation: LBP) is a feature used for classification in the field of machine vision and was proposed in 1994. . Local binary mode is a very powerful feature in the texture classification problem; if the local binary mode feature is combined with the histogram of the direction gradient, it can effectively improve the detection effect. Local binary mode is a simple but very effective texture operator. The most important property of LBP is the robustness to changes in grayscale caused by changes in lighting. Another important feature of LBP is its simple calculation, which allows it to analyze images in real time.

S3，将所述训练样本作为入参，在支持向量机SVM模型中进行SVM训练，得到图像识别模型；S3, using the training samples as input parameters, and performing SVM training in the support vector machine SVM model to obtain an image recognition model;

其中，SVM(Support Vector Machine)是一种支持向量机模型，是常见的一种判别方法。在机器学习领域，是一个有监督的学习模型，通常用来进行模式识别、分类以及回归分析。Among them, SVM (Support Vector) Machine is a support vector machine model, is a common method of discrimination. In the field of machine learning, it is a supervised learning model, usually used for pattern recognition, classification and regression analysis.

它是针对线性可分情况进行分析，对于线性不可分的情况，通过使用非线性映射算法将低维输入空间线性不可分的样本转化为高维特征空间使其线性可分，从而使得高维特征空间采用线性算法对样本的非线性特征进行线性分析成为可能；并基于结构风险最小化理论之上在特征空间中构建最优超平面，使得学习器得到全局最优化，并且在整个样本空间的期望以某个概率满足一定上界。It analyzes the linearly separable situation. For the linearly inseparable situation, the linearly separable sample in the low-dimensional input space is converted into a high-dimensional feature space by using a nonlinear mapping algorithm to make it linearly separable, so that the high-dimensional feature space is adopted. The linear algorithm makes it possible to perform linear analysis on the nonlinear characteristics of the sample; and build an optimal hyperplane in the feature space based on the structural risk minimization theory, so that the learner is globally optimized, and the expected value in the entire sample space is This probability satisfies a certain upper bound.

S4、接收上传的未知图像，根据所述图像识别模型对所述未知图像进行摩尔纹识别，以确定所述图像是否为翻拍图像。S4. Receive the uploaded unknown image, and perform moiré recognition on the unknown image according to the image recognition model to determine whether the image is a remake image.

具体的，将图像识别模型(即训练好的SVM模型)写入到SDK程序中备用，调用写入图像识别模型的SDK程序对未知图像进行截图，将截图(即未知图像)入参到SVM模型中进行摩尔纹识别，如果识别出截图(即未知图像)上带有摩尔纹则是翻拍图像，否则为真实人脸图像。Specifically, write the image recognition model (that is, the trained SVM model) to the SDK program for standby, call the SDK program that writes the image recognition model to take a screenshot of the unknown image, and enter the screenshot (that is, the unknown image) into the SVM model Moiré recognition is performed in the process, if the moiré is recognized on the screenshot (ie unknown image), it is a remake image, otherwise it is a real face image.

本实施例中，通过LBP算法来识别摩尔纹特征并经过SVM模型进行训练能够有效提升摩尔纹识别的效率，更加快速有效的对翻拍图片进行识别。In this embodiment, identifying the moiré features through the LBP algorithm and training through the SVM model can effectively improve the efficiency of moiré recognition, and more quickly and effectively identify the reprinted pictures.

图2为本申请一种基于摩尔纹的图像识别方法中的图像集形成过程示意图，如图所示，所述获取待处理的数个图像，打包成一图像集，所述图像集中包括带有摩尔纹的图像和不带有摩尔纹的图像，包括：FIG. 2 is a schematic diagram of an image set forming process in a moiré-based image recognition method of the present application. As shown in the figure, the plurality of images to be processed are acquired and packaged into an image set. Grained images and images without Moiré patterns, including:

S101、当任一所述图像到图像采集器屏幕的距离小于某一距离阈值，对图像进行采集；S101. When the distance from any of the images to the screen of the image collector is less than a certain distance threshold, collect images;

具体的，在对图像进行采集时，可以采用固定时间间隔的模式进行采集，即对需要图像采集的每一个图像进行两次拍照，取两次拍照中清晰度高的图像作为待识别图像。Specifically, when collecting images, a fixed time interval mode may be used for capturing, that is, two images are taken for each image that needs to be collected, and the high-resolution images in the two photos are taken as the image to be recognized.

两次拍照过程中转动图像采集器的角度从不同角度对所要进行摩尔纹识别的图像进行拍摄，比如，第一次采取与所要进行图像采集的图像成90°直角的方式对图像进行拍摄采集，第二次则转动图像采集器30°～45°角对所要进行摩尔纹识别的图像进行拍摄采集。采用两种不同角度的图像采集方式，可以通过不同光线入射角度对采集到的图像灰度产生影响，进而在归一化图片过程中使图片的灰度产生不同，从而防止单一角度拍摄图图像造成的摩尔纹特征不突出，影响应用LBP算法识别出摩尔纹特征的准确性。Rotate the angle of the image collector during the two photographing processes to capture the image to be moiré recognized from different angles. For example, for the first time, the image is captured at a 90 ° right angle to the image to be captured. The second time, the image collector is rotated at an angle of 30 ° to 45 ° to capture and capture the image to be identified by moiré. Using two different angles of image acquisition methods, you can affect the grayscale of the collected image through different light incident angles, and then make the grayscale of the picture different in the process of normalizing the picture, thereby preventing the image from being shot at a single angle. The moiré features are not prominent, which affects the accuracy of identifying moiré features using the LBP algorithm.

S102、将采集到的所述图像进行尺寸和色彩归一化处理，得到归一化的图像；S102. Perform normalization of the size and color of the collected image to obtain a normalized image;

具体的，在进行尺寸归一化处理时，将图像的尺寸与预设尺寸进行比较，获取该图像尺寸最进行的预设尺寸，根据此预设尺寸将图像进行拉伸或者挤压达到归一化的效果。Specifically, when the size normalization process is performed, the size of the image is compared with the preset size to obtain the preset size that is the most advanced for the image size, and the image is stretched or squeezed according to the preset size to achieve the normalization化的结果。 The effect.

S103、将所有所述归一化的图像按照生成时间进行排序，打包成一图像集。S103. Sort all the normalized images according to the generation time and package them into an image set.

具体的，赋予归一化的图像以时间标记，将同一时间生成的图像先行打包成一图像组，在建立各个图像组之间的共识节点，而后打包成一图像集。Specifically, the normalized images are given time stamps, and the images generated at the same time are packaged into an image group first, a consensus node between each image group is established, and then packaged into an image set.

本实施例中，通过对图像进行归一化处理，使得摩尔纹识别的速度和效率大大提升。In this embodiment, by normalizing the image, the speed and efficiency of moiré recognition are greatly improved.

在其中一个实施例中，所述提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本，包括：In one of the embodiments, the extracting moiré features contained in any image in the image set to form a training sample includes:

将所述图像按照水平方向和垂直方向划分为n×n等大小的子块，n为任意正整数；Divide the image into n × n equal-sized sub-blocks according to the horizontal direction and the vertical direction, where n is any positive integer;

对每个所述子块分别进行多尺度的局部二值模式LBP直方图计算，计算方法如下：Multi-scale local binary mode LBP histogram calculation is performed for each of the sub-blocks respectively, and the calculation method is as follows:

每个像素点在某尺度上的LBP _P,R值为： The LBP _{P, R} value of each pixel at a certain scale is:

其中，g _c为像素点的灰度值，g _p为以g _c为圆心，R为半径的圆周上抽取的p个像素点的灰度值，S表示影响因子，2 ^p表示模式种类数，其中p＝0，…，n； Where g _c is the gray value of the pixel, g _p is the gray value of p pixels extracted on the circumference with g _c as the center, R is the radius, S is the influence factor, and 2 ^p is the number of pattern types, Where p = 0, ..., n;

根据所述LBP _P,R值，建立所述子块的LBP直方图； Establish an LBP histogram of the sub-block according to the LBP _{P and R} values;

获取所述LBP直方图中纵坐标最大值所对应的子块，作为训练样本入参，其中，若所述LBP直方图中纵坐标最大值所对应的子块的LBP _P,R为零或者正数则说明所述图像不带有摩尔纹，为负则说明所述图像带有摩尔纹。 Obtaining the sub-block corresponding to the maximum value of the ordinate in the LBP histogram as a training sample input parameter, wherein, if the LBP _{P, R} of the sub-block corresponding to the maximum value of the ordinate in the LBP histogram is zero or positive A number indicates that the image has no moiré, and a negative value indicates that the image has moiré.

本实施例中，通过对图像进行分割成多个子块，并将每一个子块进行LBP值求解，能够更好的识别出图像中的摩尔纹特征。In this embodiment, by dividing the image into multiple sub-blocks, and solving the LBP value of each sub-block, the moiré feature in the image can be better identified.

图3为本申请一种基于摩尔纹的图像识别方法中的摩识别样本生成过程示意图，如图所示，将所述训练样本作为入参，在支持向量机SVM模型中进行SVM训练，得到图像识别模型，包括：FIG. 3 is a schematic diagram of the process of generating a friction recognition sample in a moiré-based image recognition method according to the present application. As shown in the figure, the training sample is used as an input parameter to perform SVM training in a support vector machine SVM model to obtain an image. Identify models, including:

S301、获取所述训练样本；S301: Obtain the training sample;

S302、将所述训练样本进行主成分分析PCA降维后得到低维联合特征矩阵；S302: Perform a principal component analysis and PCA dimensionality reduction on the training sample to obtain a low-dimensional joint feature matrix;

其中，PCA通常用于高维数据集的探索与可视化，还可以用于数据压缩，数据预处理等。PCA可以把可能具有相关性的高维变量合成线性无关的低维变量，称为主成分(principal components)。新的低维数据集会尽可能的保留原始数据的变量。Among them, PCA is usually used for the exploration and visualization of high-dimensional data sets, and can also be used for data compression and data preprocessing. PCA can synthesize high-dimensional variables that may have correlation into linearly independent low-dimensional variables, called principal components. The new low-dimensional data set will retain the variables of the original data as much as possible.

S303、将所述低维联合特征矩阵入参到所述SVM模型进行分类训练，得到数个两类识别器，分别为摩尔纹识别器和无摩尔纹识别器；S303. Enter the low-dimensional joint feature matrix into the SVM model for classification training to obtain several two types of recognizers, which are a moiré recognizer and a moiré-free recognizer, respectively;

其中，使用摩尔纹识别器和无摩尔纹识别器对训练样本进行分类识别，再根据识别结果进行投票，对两个类别的得票总数进行统计，票数多的类别即为该训练样本类别；采用投票的方式对样本进行统计可以减少统计过程中遗漏重要数据，如果一样多，则可以采用其它参数作为投票的依据进行投票直到选出类别。Among them, use moiré recognizer and non-moiré recognizer to classify the training samples, and then vote according to the recognition results, and count the total number of votes of the two categories. The category with the most votes is the training sample category; voting is used Statistics on the sample can reduce the missing important data in the statistical process. If there are as many as possible, other parameters can be used as the basis for voting until the category is selected.

S304、应用所述摩尔纹识别器和所述无摩尔纹识别器对所述训练样本进行摩尔纹识别，根据摩尔纹识别结果建立图像识别模型。S304. Apply the moiré identifier and the moiré-free identifier to perform moiré recognition on the training sample, and establish an image recognition model according to the moiré recognition result.

具体的，使用摩尔纹识别器或者无摩尔纹识别器对训练样本进行训练时，可以采用先使用摩尔纹识别器对每一个训练样本进行训练，识别出带有摩尔纹的图像，然后在使用无摩尔纹识别器对未识别出摩尔纹的图像进行识别；或者采用对训练样本中的每一张图片先使用摩尔纹识别器识别，再直接用无摩尔纹识别器识别，然后汇总识别结果。Specifically, when using a moiré recognizer or no moiré recognizer to train the training samples, you can first use the moiré recognizer to train each training sample to identify images with moiré, and then use the moiré recognizer The Moiré Recognizer recognizes the images that have not recognized Moiré; or it uses the Moiré Recognizer to recognize each picture in the training sample, and then directly recognizes the Moiré Recognizer, and then summarizes the recognition results.

本实施例中，通过对训练样本进行降维处理，可以提升摩尔纹识别的效率。In this embodiment, by performing dimensionality reduction processing on the training samples, the efficiency of moiré recognition can be improved.

在一个实施例中，所述提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本，包括：In one embodiment, the extracting moiré features contained in any image in the image set to form a training sample includes:

将所述图像分割成九个训练子图块，并将所述训练子图块分成二组，第一组共m块作为训练子图块，第二组共9-m块作为校验子图块，其中，1≤m≤8，且m为整数；The image is divided into nine training sub-tiles, and the training sub-tiles are divided into two groups, the first group of m blocks as training sub-tiles, and the second group of 9-m blocks as check sub-pictures Block, where 1 ≤ m ≤ 8, and m is an integer;

具体的，将图像划分成九个子图块，优选的方案是第一组为5块，第二组为4块，以便于确定摩尔纹所处的位置，对不同类型的翻拍图片摩尔纹生成位置进行位置统计，建立摩尔纹位置的线性统计数据模型，并对线性统计数据模型进行训练以便得出不同类型的翻拍图片摩尔纹出现的常见位置。在对新的图像进行识别时，可以先对某一图块所在区域进行摩尔纹识别，如果该区域识别出摩尔纹则不需要对其它区域进行识别。采用此种方式可以提升摩尔纹识别的效率，同时对于高清晰度的CCD图像采集器来说，在进行LBP特征提取像素点的过程中不需要对所有像素点进行提取节约了像素提取和灰度处理的时间。Specifically, the image is divided into nine sub-tiles, the preferred solution is that the first group is 5 blocks, and the second group is 4 blocks, so as to determine the position of the moiré, and generate the moiré for different types of reprinted pictures Perform position statistics, establish a linear statistical data model of the position of the moiré, and train the linear statistical data model in order to derive common locations where moiré appears in different types of remakes. When recognizing a new image, you can first perform moiré recognition on the area where a certain block is located. If moiré is recognized in this area, you do not need to recognize other areas. In this way, the efficiency of moiré recognition can be improved. At the same time, for high-resolution CCD image collectors, there is no need to extract all pixels in the process of LBP feature extraction, saving pixel extraction and grayscale Processing time.

提取所述训练子图块中的摩尔纹特征，其中，所述训练子图块中带有摩尔纹的训练子图块标记为“2”、不带有摩尔纹的训练子图块标记为“1”，建立训练直方图，；Extracting moiré features in the training sub-tile, where the training sub-tiles with moiré in the training sub-tile are marked as "2", and the training sub-tiles without moiré are marked as "2" 1 ", establish training histogram,

具体的，在进行直方图统计时，在直方图中，每个区域的类的数量由输入赋值栅格确定；如果指定图层，则图层的符号装置定义类的数量；如果指定数据集。Specifically, when performing histogram statistics, in the histogram, the number of classes in each area is determined by the input assignment grid; if a layer is specified, the symbol device of the layer defines the number of classes; if a data set is specified.

提取所述检验子图块中的摩尔纹特征，所述检验子图块中带有摩尔纹的检验子图块标记为“2”、不带有摩尔纹的检验子图块标记为“1”，建立检验直方图；Extracting moiré features in the inspection sub-block, the inspection sub-blocks with moiré in the inspection sub-block are marked as "2", and the inspection sub-blocks without moiré are marked as "1" , Establish inspection histogram;

汇总所述训练直方图和检验直方图得到图像直方图，若所述图像直方图上有一个子图块标记为“2”，则所述图像带有摩尔纹特征，否则所述图像不带有摩尔纹特征，并形成训练样本。Summarize the training histogram and test histogram to obtain an image histogram. If there is a sub-tile labeled "2" on the image histogram, the image has moiré features, otherwise the image does not Moiré features and form training samples.

本实施例中，对训练图块还可以进行进一步划分，即将每一个训练图块等面积划分成九个训练子图块，对每一个训练子图块进行提取LBP特征，同样建立直方图，计算统计直方图，子图块中带有摩尔纹的标记为“2”，不带有摩尔纹标记为“1”，通过统计子图块的直方图。In this embodiment, the training blocks can be further divided, that is, each training block is divided into nine training sub-blocks with equal area, and the LBP features are extracted for each training sub-block, and the histogram is also established and calculated For statistical histograms, the sub-blocks with moiré marks are marked as "2", and the moiré marks are not marked as "1". The histogram of the sub-blocks is counted.

在一个实施例中，所述将采集到的所述图像进行尺寸和色彩归一化处理，得到归一化的图像，包括：In one embodiment, the normalizing the size and color of the collected image to obtain a normalized image includes:

对所述图片库中的图片进行尺度为w、方向为h的非下采样NSCT分解，得到一个低频子图A0以及多个不同尺度和不同方向上的高频子图{A1，1；A1，2；...；Aw，1；...；Aw，h}，高频子图Aw，h描述了尺度为w，方向为h上的人脸纹理信息；Perform non-downsampling NSCT decomposition on the pictures in the picture library with a scale of w and a direction of h to obtain a low-frequency sub-picture A0 and multiple high-frequency sub-pictures of different scales and different directions {A1, 1; A1, 2; ...; Aw, 1; ...; Aw, h}, the high-frequency subgraph Aw, h describes the face texture information in the scale w and direction h;

获取各所述高频子图的尺寸，计算算数平均数得到尺寸归一化的图片；Obtain the size of each of the high-frequency sub-pictures, calculate the arithmetic mean to obtain a size-normalized picture;

将所述尺寸归一化的图片根据RGB和YUV颜色空间的变化关系所建立的亮度Y与R、G、B三个颜色分量对应关系来表达图像的灰度值，所使用的公式为：Y＝aR+bG+cB，其中a、b、c为颜色参数，0≤a、b、c≤1，即，a、b、c中任一项均为0至1之间的数，根据亮度值Y表达图像的灰度值，实现灰度的归一化。The normalized size of the picture expresses the gray value of the image according to the corresponding relationship between the brightness Y and the three color components R, G, and B established by the RGB and YUV color space. The formula used is: Y = AR + bG + cB, where a, b, and c are color parameters, and 0≤a, b, and c≤1, that is, any one of a, b, and c is a number between 0 and 1, depending on the brightness The value Y expresses the gray value of the image and realizes the normalization of gray.

具体的，在归一化处理图片过程中，可以随机提取大量的图像描述符(如SIFT、HOG等)，每个图像描述符都是一个向量，采用K-means聚类算法对这些图像描述符进行聚类，得到K个类别(K为可以调节的参数，典型值为1024、2048、10000等)。聚类中心被称为“词”，聚类得到的所有类别组成一个“码本”；Specifically, in the process of normalizing pictures, a large number of image descriptors (such as SIFT, HOG, etc.) can be randomly extracted, and each image descriptor is a vector, and these image descriptors are adopted by K-means clustering algorithm. Perform clustering to get K categories (K is an adjustable parameter, typical values are 1024, 2048, 10000, etc.). The clustering center is called "words", and all the categories obtained by clustering form a "codebook";

对于一幅图像，以稠密的方式提取特征描述符(如SIFT、HOG等)；对于每一个描述符，在码本中搜索最相似的聚类中心(也即词)。统计不同词在该图像中出现的频度，形成一个直方图。对该直方图作L1归一化，得到最后的基于词袋模型的图像纹理特征。For an image, feature descriptors (such as SIFT, HOG, etc.) are extracted in a dense manner; for each descriptor, the most similar clustering center (ie, word) is searched in the codebook. Count how often different words appear in the image to form a histogram. The histogram is normalized by L1 to obtain the final image texture feature based on the bag-of-words model.

本实施例中，通过对图片的尺寸和像素进行处理可以减少摩尔纹识别过程受到的干扰。In this embodiment, by processing the size and pixels of the picture, the interference of the moiré recognition process can be reduced.

提取所述图像的灰度I(x,y)中的二维Gabor小波纹理特征；Extract the two-dimensional Gabor wavelet texture features in the grayscale I (x, y) of the image;

对所述小波纹理特征在各尺度各方向上进行滤波卷积，Filtering and convolving the wavelet texture feature in all directions of each scale,

卷积公式为：W _u,v(x,y)＝I(x,y)×ψ _u,v(x,y)， The convolution formula is: W _{u, v} (x, y) = I (x, y) × ψ _{u, v} (x, y),

W _u,v(x,y)表示尺度v、方向u上的纹理特征向量，W _u,v的幅值包含了图像局部能量的变化，W _u,v包括Gabor核的实部和虚部响应，ψ _u,v(x,y)表示频率，I(x,y)表示灰度值； W _{u, v} (x, y) represents the texture feature vector in scale v and direction u. The amplitude of W _{u, v} contains the local energy changes of the image. W _{u, v} includes the real and imaginary parts of the Gabor kernel response , Ψ _{u, v} (x, y) represents the frequency, and I (x, y) represents the gray value;

使用FFT变换和逆FFT变换提升卷积计算的速度，公式为：Use FFT transform and inverse FFT transform to improve the speed of convolution calculation, the formula is:

W ¹ _u,v(x,y)＝F ^-1(ψ _u,v(x,y))·F(I(x,y)), W ¹ _{u, v} (x, y) = F ^-1 (ψ _{u, v} (x, y)) · F (I (x, y)),

W ¹ _u,v表示改进后的纹理特征向量，F()表示FFT变换，F ^-1()表示逆FFT变换，获得变换后的u×v个特征向量，合并这些特征向量实现图片尺寸和灰度的归一化。 W ¹ _{u, v} represents the improved texture feature vector, F () represents the FFT transform, and F ^-1 () represents the inverse FFT transform. The transformed u × v feature vectors are obtained, and these feature vectors are combined to achieve the picture size and gray Degree of normalization.

本实施例中，通过Gabor小波核函数的方法对人脸图像进行归一化处理，进一步提升摩尔纹识别的精度。In this embodiment, the face image is normalized by the Gabor wavelet kernel function method to further improve the accuracy of moiré recognition.

在一个实施例中，提出了一种基于摩尔纹的图像识别装置，如图4所示，包括如下模块：In one embodiment, an image recognition device based on moiré is proposed. As shown in FIG. 4, it includes the following modules:

在一个实施例中，所述图像集形成模块，包括：In one embodiment, the image set forming module includes:

图像采集模块，设置为当任一图像到图像采集器屏幕的距离小于某一距离阈值时，对所述图像进行采集；The image acquisition module is configured to acquire the image when the distance from any image to the image collector screen is less than a certain distance threshold;

图像归一模块，设置为将采集到的所述图像进行尺寸和色彩归一化处理，得到归一化的图像；An image normalization module, configured to normalize the size and color of the collected image to obtain a normalized image;

图像打包模块，设置为将所有所述归一化的图像按照生成时间进行排序，打包成一图像集。The image packaging module is configured to sort all the normalized images according to the generation time and package them into an image set.

在一个实施例中，所述训练样本生成，包括：In one embodiment, the training sample generation includes:

图像等分模块，设置为将所述图像按照水平方向和垂直方向划分为n×n等大小的子块，n为任意正整数；An image equalization module, configured to divide the image into n × n equal-sized sub-blocks according to the horizontal direction and the vertical direction, where n is any positive integer;

LBP特征获取模块，设置为对每个所述子块分别进行多尺度的局部二值模式LBP直方图计算，计算方法如下：The LBP feature acquisition module is set to perform multiscale local binary mode LBP histogram calculation for each of the sub-blocks respectively, and the calculation method is as follows:

直方图建立模块，设置为根据所述LBP _P,R值，建立所述子块的LBP直方图； A histogram building module, which is set to build an LBP histogram of the sub-block according to the LBP _{P and R} values;

摩尔纹识别模块，设置为获取所述LBP直方图中纵坐标最大值所对应的子块，作为训练样本入参，其中，若所述LBP直方图中纵坐标最大值所对应的子块的LBP _P,R为零或者正数则说明所述图像不带有摩尔纹，为负则说明所述图像带有摩尔纹。 The moiré recognition module is configured to obtain the sub-block corresponding to the maximum value of the ordinate in the LBP histogram as a training sample input parameter, wherein, if the LBP of the sub-block corresponding to the maximum value of the ordinate in the LBP histogram _{P, R being} zero or positive means that the image does not have moiré, and negative means that the image has moiré.

在一个实施例中，所述识别模型生成模块，包括：In one embodiment, the recognition model generation module includes:

样本获取模块，设置为获取所述训练样本；A sample acquisition module, configured to acquire the training sample;

PCA降维模块，设置为将所述训练样本进行主成分分析PCA降维后得到低维联合特征矩阵；The PCA dimensionality reduction module is configured to perform a PCA dimensionality reduction on the training samples to obtain a low-dimensional joint feature matrix;

分类训练模块，设置为将所述低维联合特征矩阵入参到所述SVM模型进行分类训练，得到数个两类识别器，分别为摩尔纹识别器和无摩尔纹识别器；The classification training module is configured to input the low-dimensional joint feature matrix into the SVM model for classification training, and obtain several two types of recognizers, namely a moiré recognizer and a non-moiré recognizer;

模型构建模块，设置为应用所述摩尔纹识别器和所述无摩尔纹识别器对所述训练样本进行摩尔纹识别，根据摩尔纹识别结果建立图像识别模型。The model building module is configured to apply the moiré recognizer and the moiré free recognizer to perform moiré recognition on the training sample, and establish an image recognition model according to the moiré recognition result.

在一个实施例中，所述未知图像识别模块，包括：In one embodiment, the unknown image recognition module includes:

图像区分模块，设置为将所述图像分割成九个训练子图块，并将所述训练子图块分成二组，第一组共m块作为训练子图块，第二组共9-m块作为校验子图块，其中，1≤m≤8，且m为整数；The image discrimination module is configured to divide the image into nine training sub-tiles, and divide the training sub-tiles into two groups, with the first group of m blocks as training sub-tiles and the second group of 9-m blocks The block serves as a parity block, where 1 ≤ m ≤ 8, and m is an integer;

训练直方图模块，设置为提取所述训练子图块中的摩尔纹特征，其中，所述训练子图块中带有摩尔纹的训练子图块标记为“2”、不带有摩尔纹的训练子图块标记为“1”，建立训练直方图；The training histogram module is set to extract the moiré features in the training sub-tiles, wherein the training sub-tiles with moiré in the training sub-tiles are marked as "2" and those without moiré The training sub-tile is marked as "1" to establish a training histogram;

检验直方图模块，设置为提取所述检验子图块中的摩尔纹特征，所述检验子图块中带有摩尔纹的检验子图块标记为“2”、不带有摩尔纹的检验子图块标记为“1”，建立检验直方图；The inspection histogram module is configured to extract the moiré features in the inspection sub-tile, the inspection sub-tiles with moiré in the inspection sub-tile are marked as "2", and the inspector without moiré The block is marked as "1" to establish the inspection histogram;

训练样本模块，设置为汇总所述训练直方图和检验直方图得到图像直方图，若所述图像直方图上有一个子图块标记为“2”，则所述图像带有摩尔纹特征，否则所述图像不带有摩尔纹特征，并形成训练样本。The training sample module is set to summarize the training histogram and the inspection histogram to obtain an image histogram. If there is a sub-tile labeled “2” on the image histogram, the image has moiré features, otherwise The image has no moiré features and forms training samples.

在一个实施例中，所述图像归一模块，包括：In one embodiment, the image normalization module includes:

纹理信息获取模块，设置为对所述图片库中的图片进行尺度为w、方向为h的非下采样NSCT分解，得到一个低频子图A0以及多个不同尺度和不同方向上的高频子图{A1，1；A1，2；...；Aw，1；...；Aw，h}，高频子图Aw，h描述了尺度为w，方向为h上的人脸纹理信息；The texture information acquisition module is set to perform non-downsampling NSCT decomposition of the pictures in the picture library with a scale of w and a direction of h to obtain a low-frequency sub-picture A0 and multiple high-frequency sub-pictures with different scales and different directions {A1, 1; A1, 2; ...; Aw, 1; ...; Aw, h}, the high-frequency subgraph Aw, h describes the face texture information on the scale w and the direction h;

图片归一化模块，设置为获取各所述高频子图的尺寸，计算算数平均数得到尺寸归一化的图片；The picture normalization module is set to obtain the size of each of the high-frequency sub-pictures, and calculate the arithmetic average to obtain a size-normalized picture;

灰度归一化模块，设置为将所述尺寸归一化的图片根据RGB和YUV颜色空间的变化关系所建立的亮度Y与R、G、B三个颜色分量对应关系来表达图像的灰度值，所使用的公式为：Y＝aR+bG+cB，其中a,b,c为颜色参数，0≤a,b,c≤1，根据亮度值Y表达图像的灰度值，实现灰度的归一化。The grayscale normalization module is set to express the grayscale of the image according to the corresponding relationship between the brightness Y and the three color components of R, G, and B established by the change relationship of the RGB and YUV color spaces. Value, the formula used is: Y = aR + bG + cB, where a, b, c are color parameters, 0≤a, b, c≤1, express the gray value of the image according to the brightness value Y, to achieve gray Normalization.

纹理提取模块，设置为提取所述图像的灰度I(x,y)中的二维Gabor小波纹理特征；The texture extraction module is set to extract the two-dimensional Gabor wavelet texture features in the grayscale I (x, y) of the image;

滤波卷积模块，设置为对所述小波纹理特征在各尺度各方向上进行滤波卷积，The filter convolution module is set to perform filter convolution on the wavelet texture features in all directions of each scale,

变换归一模块，设置为使用FFT变换和逆FFT变换提升卷积计算的速度，公式为：Transform normalization module, set to use FFT transform and inverse FFT transform to increase the speed of convolution calculation, the formula is:

在一个实施例中，提出了一种计算机设备，所述计算机设备包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行上述各实施例中的所述基于摩尔纹的图像识别方法的步骤。In one embodiment, a computer device is proposed. The computer device includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the computer device The processor executes the steps of the moiré-based image recognition method in the foregoing embodiments.

在一个实施例中，提出了一种存储有计算机可读指令的存储介质，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行上述各实施例中的所述基于摩尔纹的图像识别方法的步骤。所述存储介质可以为非易失性存储介质。In one embodiment, a storage medium storing computer-readable instructions is proposed, and when the computer-readable instructions are executed by one or more processors, the one or more processors execute the above-mentioned embodiments Steps of the image recognition method based on moiré. The storage medium may be a non-volatile storage medium.

本领域普通技术人员可以理解上述实施例的各种方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成，该程序可以存储于一计算机可读存储介质中，存储介质可以包括：只读存储器(ROM，Read Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁盘或光盘等。Those of ordinary skill in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing relevant hardware through a program. The program may be stored in a computer-readable storage medium, and the storage medium may include: Read only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, etc.

以上所述实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。The technical features of the above-mentioned embodiments can be arbitrarily combined. To simplify the description, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, All should be considered within the scope of this description.

以上所述实施例仅表达了本申请一些示例性实施例，其描述较为具体和详细，但并不能因此而理解为对本申请专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。The above-mentioned embodiments only express some exemplary embodiments of the present application, and their descriptions are more specific and detailed, but they should not be construed as limiting the patent scope of the present application. It should be noted that, for those of ordinary skill in the art, without departing from the concept of the present application, a number of modifications and improvements can also be made, which all fall within the protection scope of the present application. Therefore, the protection scope of the patent of this application shall be subject to the appended claims.

Claims

一种基于摩尔纹的图像识别方法，其中，包括：An image recognition method based on moiré, which includes:

获取待处理的数个图像，打包成一图像集，所述图像集中包括带有摩尔纹的图像和不带有摩尔纹的图像；提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本；将所述训练样本作为入参，在支持向量机SVM模型中进行SVM训练，得到图像识别模型；接收上传的未知图像，根据所述图像识别模型对所述未知图像进行摩尔纹识别，以确定所述图像是否为翻拍图像。Obtain several images to be processed and pack them into an image set, which includes images with moiré and images without moiré; extract moiré features contained in any image in the image set to form Training samples; using the training samples as input parameters, performing SVM training in a support vector machine SVM model to obtain an image recognition model; receiving an uploaded unknown image, and performing moiré recognition on the unknown image according to the image recognition model, To determine whether the image is a remake image.
根据权利要求1所述的基于摩尔纹的图像识别方法，其中，所述获取待处理的数个图像，打包成一图像集，所述图像集中包括带有摩尔纹的图像和不带有摩尔纹的图像，包括：The image recognition method based on moiré according to claim 1, wherein the acquiring several images to be processed are packaged into an image set, the image set includes images with moiré and those without moiré Images, including:

当任一图像到图像采集器屏幕的距离小于某一距离阈值时，对所述图像进行采集；When the distance between any image and the screen of the image collector is less than a certain distance threshold, the image is collected;

将采集到的所述图像进行尺寸和色彩归一化处理，得到归一化的图像；Normalize the size and color of the collected image to obtain a normalized image;

将所有所述归一化的图像按照生成时间进行排序，打包成一图像集。All the normalized images are sorted according to the generation time, and packaged into an image set.
根据权利要求1所述的基于摩尔纹的图像识别方法，其中，所述提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本，包括：The image recognition method based on moiré according to claim 1, wherein the extracting moiré features contained in any image in the image set to form a training sample includes:

将所述图像按照水平方向和垂直方向划分为n×n等大小的子块，n为任意正整数；Divide the image into n × n equal-sized sub-blocks according to the horizontal direction and the vertical direction, where n is any positive integer;

对每个所述子块分别进行多尺度的局部二值模式LBP直方图计算，计算方法如下：Multi-scale local binary mode LBP histogram calculation is performed for each of the sub-blocks respectively, and the calculation method is as follows:

每个像素点在某尺度上的LBP _P,R值为： The LBP _{P, R} value of each pixel at a certain scale is:

其中，g _c为像素点的灰度值，g _p为以g _c为圆心，R为半径的圆周上抽取的p个像素点的灰度值，S表示影响因子，2 ^p表示模式种类数，其中p＝0，…，n； Where g _c is the gray value of the pixel, g _p is the gray value of p pixels extracted on the circumference with g _c as the center, R is the radius, S is the influence factor, and 2 ^p is the number of pattern types, Where p = 0, ..., n;

根据所述LBP _P,R值，建立所述子块的LBP直方图； Establish an LBP histogram of the sub-block according to the LBP _{P and R} values;

获取所述LBP直方图中纵坐标最大值所对应的子块，作为训练样本入参，其中，若所述LBP直方图中纵坐标最大值所对应的子块的LBP _P,R为零或者正数则说明所述图像不带有摩尔纹，为负则说明所述图像带有摩尔纹。 Obtaining the sub-block corresponding to the maximum value of the ordinate in the LBP histogram as a training sample input parameter, wherein, if the LBP _{P, R} of the sub-block corresponding to the maximum value of the ordinate in the LBP histogram is zero or positive A number indicates that the image has no moiré, and a negative value indicates that the image has moiré.
根据权利要求1所述的基于摩尔纹的图像识别方法，其中，所述将所述训练样本作为入参，在支持向量机SVM模型中进行SVM训练，得到图像识别模型，包括：The image recognition method based on moiré according to claim 1, wherein the training samples are used as input parameters to perform SVM training in a support vector machine SVM model to obtain an image recognition model, including:

获取所述训练样本；Obtain the training sample;

将所述训练样本进行主成分分析PCA降维后得到低维联合特征矩阵；Principal component analysis PCA is used to reduce the dimension of the training sample to obtain a low-dimensional joint feature matrix;

将所述低维联合特征矩阵入参到所述SVM模型进行分类训练，得到数个两类识别器，分别为摩尔纹识别器和无摩尔纹识别器；The low-dimensional joint feature matrix is added to the SVM model for classification training, and several two types of recognizers are obtained, which are a moiré recognizer and a moiré-free recognizer;

应用所述摩尔纹识别器和所述无摩尔纹识别器对所述训练样本进行摩尔纹识别，根据摩尔纹识别结果建立图像识别模型。Applying the moiré recognizer and the moiré free recognizer to the moiré recognition of the training sample, and establishing an image recognition model according to the moiré recognition result.
根据权利要求1所述的基于摩尔纹的图像识别方法，其中，所述提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本，包括：The image recognition method based on moiré according to claim 1, wherein the extracting moiré features contained in any image in the image set to form a training sample includes:

将所述图像分割成九个训练子图块，并将所述训练子图块分成二组，第一组共m块作为训练子图块，第二组共9-m块作为校验子图块，其中，1≤m≤8，且m为整数；The image is divided into nine training sub-tiles, and the training sub-tiles are divided into two groups, the first group of m blocks as training sub-tiles, and the second group of 9-m blocks as check sub-pictures Block, where 1 ≤ m ≤ 8, and m is an integer;

提取所述训练子图块中的摩尔纹特征，其中，所述训练子图块中带有摩尔纹的训练子图块标记为“2”、不带有摩尔纹的训练子图块标记为“1”，建立训练直方图；Extracting moiré features in the training sub-tile, where the training sub-tiles with moiré in the training sub-tile are marked as "2", and the training sub-tiles without moiré are marked as "2" 1 ", establish training histogram;

提取所述检验子图块中的摩尔纹特征，所述检验子图块中带有摩尔纹的检验子图块标记为“2”、不带有摩尔纹的检验子图块标记为“1”，建立检验直方图；Extracting moiré features in the inspection sub-block, the inspection sub-blocks with moiré in the inspection sub-block are marked as "2", and the inspection sub-blocks without moiré are marked as "1" , Establish inspection histogram;

汇总所述训练直方图和检验直方图得到图像直方图，若所述图像直方图上有一个子图块标记为“2”，则所述图像带有摩尔纹特征，否则所述图像不带有摩尔纹特征，并形成训练样本。Summarize the training histogram and test histogram to obtain an image histogram. If there is a sub-tile labeled "2" on the image histogram, the image has moiré features, otherwise the image does not Moiré features and form training samples.
根据权利要求2所述的基于摩尔纹的图像识别方法，其中，所述将采集到的所述图像进行尺寸和色彩归一化处理，得到归一化的图像，包括：The image recognition method based on moiré according to claim 2, wherein the normalizing the size and color of the collected image to obtain a normalized image includes:

对所述图片库中的图片进行尺度为w、方向为h的非下采样NSCT分解，得到一个低频子图A0以及多个不同尺度和不同方向上的高频子图{A1，1；A1，2；...；Aw，1；...；Aw，h}，高频子图Aw，h描述了尺度为w，方向为h上的人脸纹理信息；Perform non-downsampling NSCT decomposition on the pictures in the picture library with a scale of w and a direction of h to obtain a low-frequency sub-picture A0 and multiple high-frequency sub-pictures of different scales and different directions {A1, 1; A1, 2; ...; Aw, 1; ...; Aw, h}, the high-frequency subgraph Aw, h describes the face texture information in the scale w and direction h;

获取各所述高频子图的尺寸，计算算数平均数得到尺寸归一化的图片；Obtain the size of each of the high-frequency sub-pictures, calculate the arithmetic mean to obtain a size-normalized picture;

将所述尺寸归一化的图片根据RGB和YUV颜色空间的变化关系所建立的亮度Y与R、G、B三个颜色分量对应关系来表达图像的灰度值，所使用的公式为：Y＝aR+bG+cB，其中a,b,c为颜色参数，0≤a,b,c≤1，根据亮度值Y表达图像的灰度值，实现灰度的归一化。The normalized size of the picture expresses the gray value of the image according to the corresponding relationship between the brightness Y and the three color components R, G, and B established by the RGB and YUV color space. The formula used is: Y = AR + bG + cB, where a, b, and c are color parameters, 0≤a, b, and c≤1, express the gray value of the image according to the brightness value Y, and realize the normalization of gray.
根据权利要求2所述的基于摩尔纹的图像识别方法，其中，所述将采集到的所述图像进行尺寸和色彩归一化处理，得到归一化的图像，包括：The image recognition method based on moiré according to claim 2, wherein the normalizing the size and color of the collected image to obtain a normalized image includes:

提取所述图像的灰度I(x,y)中的二维Gabor小波纹理特征；Extract the two-dimensional Gabor wavelet texture features in the grayscale I (x, y) of the image;

对所述小波纹理特征在各尺度各方向上进行滤波卷积，Filtering and convolving the wavelet texture feature in all directions of each scale,

卷积公式为：W _u,v(x,y)＝I(x,y)×ψ _u,v(x,y)， The convolution formula is: W _{u, v} (x, y) = I (x, y) × ψ _{u, v} (x, y),

W _u,v(x,y)表示尺度v、方向u上的纹理特征向量，W _u,v的幅值包含了图像局部能量的变化，W _u,v包括Gabor核的实部和虚部响应，ψ _u,v(x,y)表示频率，I(x,y)表示灰度值； W _{u, v} (x, y) represents the texture feature vector in scale v and direction u. The amplitude of W _{u, v} contains the local energy changes of the image. W _{u, v} includes the real and imaginary parts of the Gabor kernel response , Ψ _{u, v} (x, y) represents the frequency, and I (x, y) represents the gray value;

使用FFT变换和逆FFT变换提升卷积计算的速度，公式为：Use FFT transform and inverse FFT transform to improve the speed of convolution calculation, the formula is:

W ¹ _u,v(x,y)＝F ^-1(ψ _u,v(x,y))·F(I(x,y)), W ¹ _{u, v} (x, y) = F ^-1 (ψ _{u, v} (x, y)) · F (I (x, y)),

W ¹ _u,v表示改进后的纹理特征向量，F()表示FFT变换，F ^-1()表示逆FFT变换，获得变换后的u×v个特征向量，合并这些特征向量实现图片尺寸和灰度的归一化。 W ¹ _{u, v} represents the improved texture feature vector, F () represents the FFT transform, and F ^-1 () represents the inverse FFT transform. The transformed u × v feature vectors are obtained, and these feature vectors are combined to achieve the picture size and gray Degree of normalization.
一种基于摩尔纹的图像识别装置，其中，包括如下模块：An image recognition device based on moiré, which includes the following modules:

图像集形成模块，设置为获取待处理的数个图像，打包成一图像集，所述图像集中包括带有摩尔纹的图像和不带有摩尔纹的图像；The image set forming module is set to obtain several images to be processed and packaged into an image set, the image set includes images with moiré and images without moiré;

训练样本生成模块，设置为提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本；The training sample generation module is set to extract the moiré features contained in any image in the image set to form a training sample;

识别模型生成模块，设置为将所述训练样本作为入参，在支持向量机SVM模型中进行SVM训练，得到图像识别模型；The recognition model generation module is set to use the training samples as input parameters and perform SVM training in the support vector machine SVM model to obtain an image recognition model;

未知图像识别模块，设置为接收上传的未知图像，根据所述图像识别模型对所述未知图像进行摩尔纹识别，以确定所述图像是否为翻拍图像。The unknown image recognition module is configured to receive the uploaded unknown image and perform moiré recognition on the unknown image according to the image recognition model to determine whether the image is a remake image.
根据权利要求8所述的眼球运动轨迹的装置，其中，所述图像集形成模块，包括：The apparatus for eye movement track according to claim 8, wherein the image set forming module includes:

图像采集模块，设置为当任一图像到图像采集器屏幕的距离小于某一距离阈值时，对所述图像进行采集；The image acquisition module is configured to acquire the image when the distance from any image to the image collector screen is less than a certain distance threshold;

图像归一模块，设置为将采集到的所述图像进行尺寸和色彩归一化处理，得到归一化的图像；An image normalization module, configured to normalize the size and color of the collected image to obtain a normalized image;

图像打包模块，设置为将所有所述归一化的图像按照生成时间进行排序，打包成一图像集。The image packaging module is configured to sort all the normalized images according to the generation time and package them into an image set.
根据权利要求8所述的眼球运动轨迹的装置，其中，所述训练样本生成，包括：The apparatus for eye movement trajectory according to claim 8, wherein the training sample generation includes:

图像等分模块，设置为将所述图像按照水平方向和垂直方向划分为n×n等大小的子块，n为任意正整数；An image equalization module, configured to divide the image into n × n equal-sized sub-blocks according to the horizontal direction and the vertical direction, where n is any positive integer;

LBP特征获取模块，设置为对每个所述子块分别进行多尺度的局部二值模式LBP直方图计算，计算方法如下：The LBP feature acquisition module is set to perform multiscale local binary mode LBP histogram calculation for each of the sub-blocks respectively, and the calculation method is as follows:

每个像素点在某尺度上的LBP _P,R值为： The LBP _{P, R} value of each pixel at a certain scale is:

其中，g _c为像素点的灰度值，g _p为以g _c为圆心，R为半径的圆周上抽取的p个像素点的灰度值，S表示影响因子，2 ^p表示模式种类数，其中p＝0，…，n； Where g _c is the gray value of the pixel, g _p is the gray value of p pixels extracted on the circumference with g _c as the center, R is the radius, S is the influence factor, and 2 ^p is the number of pattern types, Where p = 0, ..., n;

直方图建立模块，设置为根据所述LBP _P,R值，建立所述子块的LBP直方图； A histogram building module, which is set to build an LBP histogram of the sub-block according to the LBP _{P and R} values;

摩尔纹识别模块，设置为获取所述LBP直方图中纵坐标最大值所对应的子块，作为训练样本入参，其中，若所述LBP直方图中纵坐标最大值所对应的子块的LBP _P,R为零或者正数则说明所述图像不带有摩尔纹，为负则说明所述图像带有摩尔纹。 The moiré recognition module is configured to obtain the sub-block corresponding to the maximum value of the ordinate in the LBP histogram as a training sample input parameter, wherein, if the LBP of the sub-block corresponding to the maximum value of the ordinate in the LBP histogram _{P, R being} zero or positive means that the image does not have moiré, and negative means that the image has moiré.
根据权利要求8所述的眼球运动轨迹的装置，其中，所述识别模型生成模块，包括：The device for eye movement track according to claim 8, wherein the recognition model generation module includes:

样本获取模块，设置为获取所述训练样本；A sample acquisition module, configured to acquire the training sample;

PCA降维模块，设置为将所述训练样本进行主成分分析PCA降维后得到低维联合特征矩阵；The PCA dimensionality reduction module is configured to perform a PCA dimensionality reduction on the training samples to obtain a low-dimensional joint feature matrix;

分类训练模块，设置为将所述低维联合特征矩阵入参到所述SVM模型进行分类训练，得到数个两类识别器，分别为摩尔纹识别器和无摩尔纹识别器；The classification training module is configured to input the low-dimensional joint feature matrix into the SVM model for classification training, and obtain several two types of recognizers, namely a moiré recognizer and a non-moiré recognizer;

模型构建模块，设置为应用所述摩尔纹识别器和所述无摩尔纹识别器对所述训练样本进行摩尔纹识别，根据摩尔纹识别结果建立图像识别模型。The model building module is configured to apply the moiré recognizer and the moiré free recognizer to perform moiré recognition on the training sample, and establish an image recognition model according to the moiré recognition result.
根据权利要求8所述的眼球运动轨迹的装置，其中，所述未知图像识别模块，包括：The device for eye movement trajectory according to claim 8, wherein the unknown image recognition module includes:

图像区分模块，设置为将所述图像分割成九个训练子图块，并将所述训练子图块分成二组，第一组共m块作为训练子图块，第二组共9-m块作为校验子图块，其中，1≤m≤8，且m为整数；The image discrimination module is configured to divide the image into nine training sub-tiles, and divide the training sub-tiles into two groups, with the first group of m blocks as training sub-tiles and the second group of 9-m blocks The block serves as a parity block, where 1 ≤ m ≤ 8, and m is an integer;

训练直方图模块，设置为提取所述训练子图块中的摩尔纹特征，其中，所述训练子图块中带有摩尔纹的训练子图块标记为“2”、不带有摩尔纹的训练子图块标记为“1”，建立训练直方图；The training histogram module is set to extract the moiré features in the training sub-tiles, wherein the training sub-tiles with moiré in the training sub-tiles are marked as "2" and those without moiré The training sub-tile is marked as "1" to establish a training histogram;

检验直方图模块，设置为提取所述检验子图块中的摩尔纹特征，所述检验子图块中带有摩尔纹的检验子图块标记为“2”、不带有摩尔纹的检验子图块标记为“1”，建立检验直方图；The inspection histogram module is configured to extract the moiré features in the inspection sub-tile, the inspection sub-tiles with moiré in the inspection sub-tile are marked as "2", and the inspector without moiré The block is marked as "1" to establish the inspection histogram;

训练样本模块，设置为汇总所述训练直方图和检验直方图得到图像直方图，若所述图像直方图上有一个子图块标记为“2”，则所述图像带有摩尔纹特征，否则所述图像不带有摩尔纹特征，并形成训练样本。The training sample module is set to summarize the training histogram and the inspection histogram to obtain an image histogram. If there is a sub-tile labeled “2” on the image histogram, the image has moiré features, otherwise The image has no moiré features and forms training samples.
根据权利要求9所述的眼球运动轨迹的装置，其中，所述图像归一模块，包括：The device for eye movement trajectory according to claim 9, wherein the image normalization module includes:

纹理信息获取模块，设置为对所述图片库中的图片进行尺度为w、方向为h的非下采样NSCT分解，得到一个低频子图A0以及多个不同尺度和不同方向上的高频子图{A1，1；A1，2；...；Aw，1；...；Aw，h}，高频子图Aw，h描述了尺度为w，方向为h上的人脸纹理信息；The texture information acquisition module is set to perform non-downsampling NSCT decomposition of the pictures in the picture library with a scale of w and a direction of h to obtain a low-frequency sub-picture A0 and multiple high-frequency sub-pictures with different scales and different directions {A1, 1; A1, 2; ...; Aw, 1; ...; Aw, h}, the high-frequency subgraph Aw, h describes the face texture information on the scale w and the direction h;

图片归一化模块，设置为获取各所述高频子图的尺寸，计算算数平均数得到尺寸归一化的图片；The picture normalization module is set to obtain the size of each of the high-frequency sub-pictures, and calculate the arithmetic average to obtain a size-normalized picture;

灰度归一化模块，设置为将所述尺寸归一化的图片根据RGB和YUV颜色空间的变化关系所建立的亮度Y与R、G、B三个颜色分量对应关系来表达图像的灰度值，所使用的公式为：Y＝aR+bG+cB，其中a,b,c为颜色参数，0≤a,b,c≤1，根据亮度值Y表达图像的灰度值，实现灰度的归一化。The grayscale normalization module is set to express the grayscale of the image according to the corresponding relationship between the brightness Y and the three color components of R, G, and B established by the change relationship of the RGB and YUV color spaces. Value, the formula used is: Y = aR + bG + cB, where a, b, c are color parameters, 0≤a, b, c≤1, express the gray value of the image according to the brightness value Y, to achieve gray Normalization.
根据权利要求9所述的眼球运动轨迹的装置，其中，所述图像归一模块，包括：The device for eye movement trajectory according to claim 9, wherein the image normalization module includes:

纹理提取模块，设置为提取所述图像的灰度I(x,y)中的二维Gabor小波纹理特征；The texture extraction module is set to extract the two-dimensional Gabor wavelet texture features in the grayscale I (x, y) of the image;

滤波卷积模块，设置为对所述小波纹理特征在各尺度各方向上进行滤波卷积，The filter convolution module is set to perform filter convolution on the wavelet texture features in all directions of each scale,

卷积公式为：W _u,v(x,y)＝I(x,y)×ψ _u,v(x,y)， The convolution formula is: W _{u, v} (x, y) = I (x, y) × ψ _{u, v} (x, y),

W _u,v(x,y)表示尺度v、方向u上的纹理特征向量，W _u,v的幅值包含了图像局部能量的变化，W _u,v包括Gabor核的实部和虚部响应，ψ _u,v(x,y)表示频率，I(x,y)表示灰度值； W _{u, v} (x, y) represents the texture feature vector in scale v and direction u. The amplitude of W _{u, v} contains the local energy changes of the image. W _{u, v} includes the real and imaginary parts of the Gabor kernel response , Ψ _{u, v} (x, y) represents the frequency, and I (x, y) represents the gray value;

变换归一模块，设置为使用FFT变换和逆FFT变换提升卷积计算的速度，公式为：Transform normalization module, set to use FFT transform and inverse FFT transform to increase the speed of convolution calculation, the formula is:

W ¹ _u,v(x,y)＝F ^-1(ψ _u,v(x,y))·F(I(x,y)), W ¹ _{u, v} (x, y) = F ^-1 (ψ _{u, v} (x, y)) · F (I (x, y)),

W ¹ _u,v表示改进后的纹理特征向量，F()表示FFT变换，F ^-1()表示逆FFT变换，获得变换后的u×v个特征向量，合并这些特征向量实现图片尺寸和灰度的归一化。 W ¹ _{u, v} represents the improved texture feature vector, F () represents the FFT transform, and F ^-1 () represents the inverse FFT transform. The transformed u × v feature vectors are obtained, and these feature vectors are combined to achieve the picture size and gray Degree of normalization.
一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行以下步骤：A computer device includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor causes the processor to perform the following steps:

获取待处理的数个图像，打包成一图像集，所述图像集中包括带有摩尔纹的图像和不带有摩尔纹的图像；Obtain several images to be processed and pack them into an image set, where the image set includes images with moiré and images without moiré;

提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本；Extracting moiré features contained in any image in the image set to form a training sample;

将所述训练样本作为入参，在支持向量机SVM模型中进行SVM训练，得到图像识别模型；Using the training samples as input parameters, performing SVM training in the support vector machine SVM model to obtain an image recognition model;

接收上传的未知图像，根据所述图像识别模型对所述未知图像进行摩尔纹识别，以确定所述图像是否为翻拍图像。Receiving the uploaded unknown image, and performing moiré recognition on the unknown image according to the image recognition model to determine whether the image is a remake image.
一种存储有计算机可读指令的存储介质，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行以下步骤：A storage medium storing computer-readable instructions, which when executed by one or more processors, causes the one or more processors to perform the following steps:

获取待处理的数个图像，打包成一图像集，所述图像集中包括带有摩尔纹的图像和不带有摩尔纹的图像；提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本；将所述训练样本作为入参，在支持向量机SVM模型中进行SVM训练，得到图像识别模型；接收上传的未知图像，根据所述图像识别模型对所述未知图像进行摩尔纹识别，以确定所述图像是否为翻拍图像。Obtain several images to be processed and pack them into an image set, which includes images with moiré and images without moiré; extract moiré features contained in any image in the image set to form Training samples; using the training samples as input parameters, performing SVM training in a support vector machine SVM model to obtain an image recognition model; receiving an uploaded unknown image, and performing moiré recognition on the unknown image according to the image recognition model, To determine whether the image is a remake image.
根据权利要求16所述的一种存储有计算机可读指令的存储介质，其中，所述获取待处理的数个图像，打包成一图像集，所述图像集中包括带有摩尔纹的图像和不带有摩尔纹的图像时，使得所述处理器执行以下步骤：A storage medium storing computer readable instructions according to claim 16, wherein the acquisition of several images to be processed is packaged into an image set, and the image set includes images with moiré and without When there are moiré images, the processor is allowed to perform the following steps:

当任一图像到图像采集器屏幕的距离小于某一距离阈值时，对所述图像进行采集；When the distance between any image and the screen of the image collector is less than a certain distance threshold, the image is collected;

将采集到的所述图像进行尺寸和色彩归一化处理，得到归一化的图像；Normalize the size and color of the collected image to obtain a normalized image;

将所有所述归一化的图像按照生成时间进行排序，打包成一图像集。All the normalized images are sorted according to the generation time, and packaged into an image set.
根据权利要求16所述的一种存储有计算机可读指令的存储介质，其中，所述提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本时，使得所述处理器执行以下步骤：A storage medium storing computer readable instructions according to claim 16, wherein said extracting the moiré features contained in any image in the image set to form a training sample causes the processor to execute The following steps:

将所述图像按照水平方向和垂直方向划分为n×n等大小的子块，n为任意正整数；Divide the image into n × n equal-sized sub-blocks according to the horizontal direction and the vertical direction, where n is any positive integer;

对每个所述子块分别进行多尺度的局部二值模式LBP直方图计算，计算方法如下：Multi-scale local binary mode LBP histogram calculation is performed for each of the sub-blocks respectively, and the calculation method is as follows:

每个像素点在某尺度上的LBP _P,R值为： The LBP _{P, R} value of each pixel at a certain scale is:

其中，g _c为像素点的灰度值，g _p为以g _c为圆心，R为半径的圆周上抽取的p个像素点的灰度值，S表示影响因子，2 ^p表示模式种类数，其中p＝0，…，n； Where g _c is the gray value of the pixel, g _p is the gray value of p pixels extracted on the circumference with g _c as the center, R is the radius, S is the influence factor, and 2 ^p is the number of pattern types, Where p = 0, ..., n;

根据所述LBP _P,R值，建立所述子块的LBP直方图； Establish an LBP histogram of the sub-block according to the LBP _{P and R} values;

获取所述LBP直方图中纵坐标最大值所对应的子块，作为训练样本入参，其中，若所述LBP直方图中纵坐标最大值所对应的子块的LBP _P,R为零或者正数则说明所述图像不带有摩尔纹，为负则说明所述图像带有摩尔纹。 Obtaining the sub-block corresponding to the maximum value of the ordinate in the LBP histogram as a training sample input parameter, wherein, if the LBP _{P, R} of the sub-block corresponding to the maximum value of the ordinate in the LBP histogram is zero or positive A number indicates that the image has no moiré, and a negative value indicates that the image has moiré.
根据权利要求16所述的一种存储有计算机可读指令的存储介质，其中，所述将所述训练样本作为入参，在支持向量机SVM模型中进行SVM训练，得到图像识别模型时，使得所述处理器执行以下步骤：A storage medium storing computer readable instructions according to claim 16, wherein the training samples are used as input parameters to perform SVM training in a support vector machine SVM model to obtain an image recognition model such that The processor performs the following steps:

获取所述训练样本；将所述训练样本进行主成分分析PCA降维后得到低维联合特征矩阵；将所述低维联合特征矩阵入参到所述SVM模型进行分类训练，得到数个两类识别器，分别为摩尔纹识别器和无摩尔纹识别器；应用所述摩尔纹识别器和所述无摩尔纹识别器对所述训练样本进行摩尔纹识别，根据摩尔纹识别结果建立图像识别模型。Obtaining the training samples; subjecting the training samples to PCA dimensionality reduction to obtain a low-dimensional joint feature matrix; adding the low-dimensional joint feature matrix to the SVM model for classification training to obtain several two categories Recognizers, respectively, moiré recognizer and non-moiré recognizer; apply the moiré recognizer and the moiré free recognizer to the moiré recognition of the training sample, and establish an image recognition model according to the moiré recognition result .
根据权利要求16所述的一种存储有计算机可读指令的存储介质，其中，所述提取所述图像集中的任一图像所包含的摩尔纹特征，形成训练样本时，使得所述处理器执行以下步骤：A storage medium storing computer readable instructions according to claim 16, wherein said extracting the moiré features contained in any image in the image set to form a training sample causes the processor to execute The following steps:

将所述图像分割成九个训练子图块，并将所述训练子图块分成二组，第一组共m块作为训练子图块，第二组共9-m块作为校验子图块，其中，1≤m≤8，且m为整数；The image is divided into nine training sub-tiles, and the training sub-tiles are divided into two groups, the first group of m blocks as training sub-tiles, and the second group of 9-m blocks as check sub-pictures Block, where 1 ≤ m ≤ 8, and m is an integer;

提取所述训练子图块中的摩尔纹特征，其中，所述训练子图块中带有摩尔纹的训练子图块标记为“2”、不带有摩尔纹的训练子图块标记为“1”，建立训练直方图；Extracting moiré features in the training sub-tile, where the training sub-tiles with moiré in the training sub-tile are marked as "2", and the training sub-tiles without moiré are marked as "2" 1 ", establish training histogram;

提取所述检验子图块中的摩尔纹特征，所述检验子图块中带有摩尔纹的检验子图块标记为“2”、不带有摩尔纹的检验子图块标记为“1”，建立检验直方图；Extracting moiré features in the inspection sub-block, the inspection sub-blocks with moiré in the inspection sub-block are marked as "2", and the inspection sub-blocks without moiré are marked as "1" , Establish inspection histogram;

汇总所述训练直方图和检验直方图得到图像直方图，若所述图像直方图上有一个子图块标记为“2”，则所述图像带有摩尔纹特征，否则所述图像不带有摩尔纹特征，并形成训练样本。Summarize the training histogram and test histogram to obtain an image histogram. If there is a sub-tile labeled "2" on the image histogram, the image has moiré features, otherwise the image does not Moiré features and form training samples.