WO2021031446A1 - 一种基于二维动态特征的离线笔迹个体识别系统及方法 - Google Patents

一种基于二维动态特征的离线笔迹个体识别系统及方法 Download PDF

Info

Publication number
WO2021031446A1
WO2021031446A1 PCT/CN2019/122178 CN2019122178W WO2021031446A1 WO 2021031446 A1 WO2021031446 A1 WO 2021031446A1 CN 2019122178 W CN2019122178 W CN 2019122178W WO 2021031446 A1 WO2021031446 A1 WO 2021031446A1
Authority
WO
WIPO (PCT)
Prior art keywords
handwriting
suspicious
image
offline
feature
Prior art date
Application number
PCT/CN2019/122178
Other languages
English (en)
French (fr)
Inventor
陈晓红
杨旭
王雅晨
王楠
卢启萌
Original Assignee
司法鉴定科学研究院
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
Priority to US16/969,492 priority Critical patent/US11854284B2/en
Application filed by 司法鉴定科学研究院 filed Critical 司法鉴定科学研究院
Publication of WO2021031446A1 publication Critical patent/WO2021031446A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/32—Digital ink
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/32—Digital ink
    • G06V30/36—Matching; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/16—Image preprocessing
    • G06V30/168—Smoothing or thinning of the pattern; Skeletonisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/32—Digital ink
    • G06V30/333—Preprocessing; Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/32—Digital ink
    • G06V30/333—Preprocessing; Feature extraction
    • G06V30/347—Sampling; Contour coding; Stroke extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/30—Writer recognition; Reading and verifying signatures

Definitions

  • the invention relates to the technical field of identity recognition based on behavior characteristics, in particular to an offline handwriting individual recognition system and method based on two-dimensional dynamic characteristics.
  • the existing handwriting recognition technology is divided into online handwriting recognition and offline handwriting recognition.
  • online handwriting recognition can obtain more information about writing sequence, speed, pressure, angle and other characteristics that are beneficial to identity recognition in real time, but this information needs to be obtained by special equipment, thus limiting its application scope and development prospects;
  • Offline handwriting is the trajectory of writing motion, that is, displayed in front of people in the form of a two-dimensional static image, so offline handwriting recognition needs to mine and extract feature information from it.
  • the accuracy of offline handwriting recognition is not high.
  • the present invention provides an offline handwriting individual recognition system based on two-dimensional dynamic features, which specifically includes:
  • a data acquisition module for acquiring a first handwriting image corresponding to suspicious handwriting, and acquiring a second handwriting image corresponding to a number of sample handwriting;
  • the data preprocessing module connected to the data acquisition module, is used to perform image preprocessing on the first handwriting image to obtain a first preprocessing image corresponding to the suspicious handwriting, and to perform all the second handwriting images separately The image preprocessing to obtain a second preprocessed image corresponding to each of the sample handwriting;
  • the skeleton extraction module connected to the data preprocessing module, is used to perform skeleton extraction on the first preprocessed image to obtain the first skeleton image corresponding to the suspicious handwriting, and perform the respective processing on each of the second preprocessed images. Extracting the skeleton to obtain a second skeleton image corresponding to the sample handwriting;
  • the line width of the first skeleton image and each of the second skeleton images is a single pixel
  • the stroke order recognition module connected to the skeleton extraction module, is configured to perform handwriting tracking according to the first skeleton image to obtain the first writing trajectory corresponding to the suspicious handwriting, and to perform the handwriting tracking respectively according to each of the second skeleton images Obtaining a second writing track corresponding to each of the sample handwriting;
  • the feature extraction module is used to perform feature extraction on the corresponding first handwriting image according to the first writing trajectory to obtain each pixel in the first writing trajectory A first dynamic feature of a point, and extracting a second dynamic feature of each pixel in each of the second writing trajectories from the corresponding second handwriting images according to each of the second writing trajectories;
  • the first data processing module connected to the feature extraction module, is configured to process the suspicious handwriting according to the first dynamic feature corresponding to the suspicious handwriting and the second dynamic feature corresponding to each sample handwriting The correlation coefficient with each of the sample handwriting;
  • the second data processing module connected to the first data processing module, is used to process the data of the suspicious handwriting based on the first probability density distribution data and the second probability density distribution data obtained by statistics in advance, and each of the correlation coefficients. Individual identification results.
  • the first data processing module specifically includes:
  • a feature pair generating unit configured to combine the first dynamic feature and each of the second dynamic features pairwise to generate multiple feature pairs
  • the dynamic time warping unit is connected to the feature pair generating unit, and is used to perform dynamic time warping between the first dynamic feature and the second dynamic feature in each feature pair, so that the first dynamic feature Match the strokes corresponding to the second dynamic feature and the corresponding parts corresponding to the strokes one by one;
  • the data calculation unit is connected to the dynamic time warping unit, and is used to respectively calculate the correlation coefficient between the first dynamic feature and the second dynamic feature in each of the feature pairs after the dynamic time warping is performed.
  • the second data processing module specifically includes:
  • the first data processing unit is configured to perform average processing on all the correlation coefficients corresponding to the suspicious handwriting to obtain the mean correlation coefficient corresponding to the suspicious handwriting;
  • the second data processing unit connected to the first data processing unit, is used to calculate the first probability corresponding to the mean correlation coefficient in the first probability density distribution data, and to calculate the mean correlation coefficient in the first probability density distribution data. 2.
  • the first probability is the probability that the suspicious handwriting and the sample handwriting corresponding to the mean correlation coefficient are written by the same person
  • the second probability is the probability that the suspicious handwriting and the sample handwriting corresponding to the mean correlation coefficient are Probability that the sample handwriting is not written by the same person
  • a third data processing unit connected to the second data processing unit, and configured to calculate the ratio of the first probability to the second probability to obtain the likelihood ratio of the suspicious handwriting;
  • the result judging module is connected to the third data processing unit, and is used for matching with a preset likelihood ratio scale according to the likelihood ratio to obtain the individual recognition result of the suspicious handwriting.
  • the first probability density distribution data is the probability density distribution of the correlation coefficient between the handwritings of several real samples of the person written by the same person, and
  • the second probability density distribution data is the probability density distribution of the correlation coefficients between the handwriting of several other people and the handwriting of the real sample obtained by different people according to the handwriting of the real sample of the person.
  • the handwriting of the other person includes handwriting for signature, and/or copying handwriting, and/or copying handwriting.
  • the image preprocessing method includes removing the background, and/or filling the whiteness in the strokes, and/or smoothing the strokes, and/or binarization.
  • the first dynamic feature includes width, and/or gray scale, and/or radian.
  • the second dynamic feature includes width, and/or gray scale, and/or radian.
  • An offline handwriting individual recognition method based on two-dimensional dynamic features which is applied to the offline handwriting individual recognition system described in any one of the above, specifically includes the following steps:
  • Step S1 the offline handwriting individual recognition system obtains a first handwriting image corresponding to a suspicious handwriting, and obtains a second handwriting image corresponding to a number of sample handwritings;
  • Step S2 the offline handwriting individual recognition system performs image preprocessing on the first handwriting image to obtain a first preprocessed image corresponding to the suspicious handwriting, and performs the image preprocessing on each of the second handwriting images respectively Obtaining a second preprocessed image corresponding to each of the sample handwriting;
  • Step S3 the offline handwriting individual recognition system performs skeleton extraction on the first preprocessed image to obtain a first skeleton image corresponding to the suspicious handwriting, and performs the skeleton extraction on each of the second preprocessed images respectively The second skeleton image corresponding to the sample handwriting;
  • the line width of the first skeleton image and each second skeleton image is a single pixel
  • Step S4 the offline handwriting individual recognition system separately performs handwriting tracking according to the first skeleton image to obtain the first writing trajectory corresponding to the suspicious handwriting, and performs the handwriting tracking separately according to each of the second skeleton images to obtain each The second writing track corresponding to the sample handwriting;
  • Step S5 the offline handwriting recognition system performs feature extraction on the corresponding first handwriting image according to the first writing trajectory to obtain the first dynamic feature of each pixel in the first writing trajectory, and Extracting from each corresponding second handwriting image according to each second writing trajectory to obtain a second dynamic feature of each pixel in each second writing trajectory;
  • Step S6 the offline handwriting individual recognition system processes the suspicious handwriting and each sample according to the first dynamic feature corresponding to the suspicious handwriting and the second dynamic feature corresponding to each sample handwriting Correlation coefficient between handwriting;
  • step S7 the offline handwriting individual recognition system processes and obtains the individual recognition result of the suspicious handwriting based on the first probability density distribution data and the second probability density distribution data obtained by statistics in advance, and the respective correlation coefficients.
  • the step S7 specifically includes:
  • Step S71 the offline handwriting individual recognition system combines the first dynamic feature and each of the second dynamic features to generate multiple feature pairs;
  • Step S72 the offline handwriting individual recognition system performs dynamic time warping between the first dynamic feature and the second dynamic feature in each of the feature pairs, so that the first dynamic feature and the first dynamic feature 2.
  • the strokes corresponding to the dynamic features are matched with the corresponding parts of the strokes one by one;
  • step S73 the offline handwriting individual recognition system respectively calculates the correlation coefficient between the first dynamic feature and the second dynamic feature in each of the feature pairs after the dynamic time warping is performed.
  • the step S8 specifically includes:
  • Step S81 the offline handwriting individual recognition system averages all the correlation coefficients corresponding to the suspicious handwriting to obtain the mean correlation coefficient corresponding to the suspicious handwriting;
  • Step S82 the offline handwriting individual recognition system calculates the first probability corresponding to the mean correlation coefficient in the first probability density distribution data, and calculates the mean correlation coefficient corresponding to the second probability density distribution data Second probability
  • the first probability is the probability that the suspicious handwriting and the sample handwriting corresponding to the mean correlation coefficient are written by the same person
  • the second probability is the probability that the suspicious handwriting and the sample handwriting corresponding to the mean correlation coefficient are Probability that the sample handwriting is not written by the same person
  • Step S83 the offline handwriting individual recognition system calculates the ratio of the first probability to the second probability to obtain the likelihood ratio of the suspicious handwriting
  • step S84 the offline handwriting individual recognition system performs matching with a preset likelihood ratio scale according to the likelihood ratio to obtain an individual recognition result of the suspicious handwriting.
  • the handwriting images of suspicious handwriting and sample handwriting can be obtained only by two-dimensional scanning equipment, and the image preprocessing technology can effectively remove the interference of factors such as paper background, adapt to the writing results of different thickness writing tools, and enhance the detection of different Object compatibility;
  • FIG. 1 is a schematic structural diagram of an offline handwriting individual recognition system based on two-dimensional dynamic features in a preferred embodiment of the present invention
  • Figure 2 is a schematic diagram of a likelihood ratio scale in a preferred embodiment of the present invention.
  • FIG. 3 is a schematic flowchart of a method for offline handwriting individual recognition based on two-dimensional dynamic features in a preferred embodiment of the present invention
  • FIG. 4 is a schematic diagram of a sub-process of an offline handwriting individual recognition method based on two-dimensional dynamic features in a preferred embodiment of the present invention
  • FIG. 5 is a schematic diagram of a sub-process of an offline handwriting individual recognition method based on two-dimensional dynamic features in a preferred embodiment of the present invention.
  • FIG. 1 specifically includes:
  • the data acquisition module 1 is used to acquire a first handwriting image corresponding to suspicious handwriting, and to acquire a second handwriting image corresponding to several sample handwritings;
  • Data preprocessing module 2 connected to data acquisition module 1, for image preprocessing the first handwriting image to obtain the first preprocessing image corresponding to the suspicious handwriting, and performing image preprocessing on each second handwriting image to obtain each sample handwriting The corresponding second preprocessed image;
  • the skeleton extraction module 3 connected to the data preprocessing module 2, is used to perform skeleton extraction on the first preprocessed image to obtain the first skeleton image corresponding to the suspicious handwriting, and perform skeleton extraction on each second preprocessed image to obtain the corresponding sample handwriting.
  • Second skeleton image is used to perform skeleton extraction on the first preprocessed image to obtain the first skeleton image corresponding to the suspicious handwriting, and perform skeleton extraction on each second preprocessed image to obtain the corresponding sample handwriting.
  • the line width of the first skeleton image and each second skeleton image is a single pixel
  • the stroke order recognition module 4 connected to the skeleton extraction module 3, is used to track the handwriting according to the first skeleton image to obtain the first writing trajectory corresponding to the suspicious handwriting, and to separately track the handwriting according to each second skeleton image to obtain the second handwriting corresponding to each sample Writing track
  • the feature extraction module 5 is used to perform feature extraction on the corresponding first handwriting image according to the first writing trajectory to obtain the first dynamics of each pixel in the first writing trajectory Feature, and extract the second dynamic feature of each pixel in each second writing trajectory separately from each corresponding second handwriting image according to each second writing trajectory;
  • the first data processing module 6, connected to the feature extraction module 5, is used for processing to obtain the correlation coefficient between the suspicious handwriting and each sample handwriting according to the first dynamic feature corresponding to the suspicious handwriting and the second dynamic feature corresponding to each sample handwriting;
  • the second data processing module 7 is connected to the first data processing module 6, and is used to process the individual recognition results of the suspicious handwriting according to the first probability density distribution data and the second probability density distribution data obtained by statistics in advance, and the correlation coefficients.
  • the computer connected to the large-area optical two-dimensional profile scanning instrument obtains the first handwriting image corresponding to the suspicious handwriting and the second handwriting image corresponding to each sample handwriting.
  • the first handwriting image and the second handwriting image are respectively subjected to image preprocessing to obtain a first preprocessed image corresponding to the first handwriting image, and a second preprocessed image corresponding to the second handwriting image
  • the above image preprocessing includes removing Operations such as background, filling in whiteness in strokes, smoothing strokes and binarization.
  • the available handwriting and sample handwriting to be detected can also be photographed by a camera, thereby obtaining the first handwriting image corresponding to the suspicious handwriting and obtaining the second handwriting image corresponding to each sample handwriting.
  • the processing procedures for suspicious handwriting and sample handwriting are similar. Therefore, only the processing procedure of suspicious handwriting is taken as an example for detailed description, and the processing procedure for sample handwriting is not repeated here.
  • the process of handling suspicious handwriting is as follows:
  • the skeleton of the corresponding suspicious handwriting is extracted from the first preprocessed image to obtain the first skeleton image corresponding to the suspicious handwriting, and the first writing track corresponding to the suspicious handwriting is obtained by tracking the handwriting on the first skeleton image.
  • the above-mentioned handwriting tracking is completed in the form of human-computer interaction, which specifically includes:
  • the first dynamic feature of each pixel on the first writing trajectory is acquired on the corresponding first handwriting image, that is, the characteristics of stroke width, stroke gray scale, and stroke radian.
  • the dynamic feature is the two-dimensional dynamic feature of the suspicious handwriting.
  • the sample handwriting is processed according to the same processing procedure described above to obtain the two-dimensional dynamic characteristics of the sample handwriting. Then the correlation coefficient between the two-dimensional dynamic feature of the suspicious handwriting and the two-dimensional dynamic feature of the sample handwriting is calculated, and the similarity between the suspicious handwriting and the sample handwriting is characterized by the correlation coefficient. Specifically, because the formation of handwriting is a process, even if the handwriting is written by the same person, because the writing speed is not exactly the same each time, the length of each corresponding stroke is different, resulting in different sequence lengths corresponding to the handwriting, and the same stroke occurs in the sequence. dislocation.
  • the sequence length of each stroke when the two-dimensional dynamic feature of the suspicious handwriting is projected on the above-mentioned coordinate system is different from the two-dimensional dynamic feature of the sample handwriting when projected on the above-mentioned coordinate system.
  • the sequence lengths of the strokes are not exactly the same, resulting in the inability to calculate the effective correlation coefficient between the two-dimensional dynamic feature of the suspicious handwriting and the two-dimensional dynamic feature of the sample handwriting. Therefore, before calculating the correlation coefficient, it is necessary to perform dynamic time warping on the two-dimensional dynamic features of a set of suspicious handwriting and the two-dimensional dynamic features of the sample handwriting that need to be calculated for the correlation coefficient.
  • DTW technology is used for dynamic time warping.
  • the above-mentioned similarity that is, the correlation coefficient
  • the correlation coefficient needs to be substituted into the probability density distribution obtained by statistics in advance for further processing.
  • the result of the comparison has only one correlation coefficient, which is the mean correlation coefficient described in the present invention.
  • handwriting recognition is usually performed, several suspicious handwritings and several sample handwritings are obtained. For example, two suspicious handwritings and three sample handwritings are obtained by combining the two-dimensional dynamic characteristics of the two suspicious handwritings with the three samples. The two-dimensional dynamic characteristics of the handwriting are calculated for the correlation coefficient, and for each suspicious handwriting, the calculation results of the corresponding three correlation coefficients will be obtained.
  • the first probability density distribution data obtained by the foregoing statistics is the probability density distribution of the correlation coefficients between several real sample handwritings of the same person written by the same person
  • the second probability density distribution data is the probability density distribution of the correlation coefficients between different people according to the above-mentioned person.
  • the probability density distribution of the correlation coefficients between the handwritings of a number of other persons obtained from the handwriting of the real sample and the handwriting of the above-mentioned real sample.
  • hundreds of thousands of handwriting data are obtained in advance to form a corresponding handwriting database.
  • the handwriting database includes a number of real sample handwritings written by the same person, that is, the handwriting of the person, and also includes handwriting written by different people based on the real sample of the person.
  • the handwritings obtained by others are the handwritings obtained from the above-mentioned personal handwritings, and/or copying handwritings, and/or copying handwritings, etc.
  • the likelihood ratio of suspicious handwriting is further calculated by the following formula:
  • SLR means the likelihood ratio of suspicious handwriting
  • H p means that the suspicious handwriting and the sample handwriting are written by the same person
  • H p means that the suspicious handwriting and the sample handwriting are not written by the same person
  • E U means the suspicious handwriting
  • E S means the sample handwriting ;
  • a likelihood ratio scale is set in advance, and the calculation result of the likelihood ratio is matched with the likelihood ratio scale. If the likelihood ratio is greater than 10,000, the individual The recognition result strongly supports that the suspicious handwriting and the sample handwriting are written by the same person; if the likelihood ratio is less than 0.001, the individual recognition result given strongly supports that the suspicious handwriting and the sample handwriting are not written by the same person; other matches The result can be deduced by analogy.
  • the first data processing module 6 specifically includes:
  • the feature pair generating unit 61 is configured to combine the first dynamic feature and each second dynamic feature in pairs to generate multiple feature pairs;
  • the dynamic time warping unit 62 which connects the feature pair generating unit 61, is used for dynamic time warping between the first dynamic feature and the second dynamic feature in each feature pair, so that the first dynamic feature and the second dynamic feature correspond to the strokes Match the corresponding parts of the strokes one by one;
  • the data calculation unit 63 is connected to the dynamic time warping unit 62, and is used to calculate the correlation coefficients between the first dynamic feature and the second dynamic feature in each feature pair after dynamic time warping.
  • the second data processing module 7 specifically includes:
  • the first data processing unit 71 is configured to perform average processing on all correlation coefficients corresponding to the suspicious handwriting to obtain the mean correlation coefficient corresponding to the suspicious handwriting;
  • the second data processing unit 72 connected to the first data processing unit 71, is used to calculate the first probability of the mean correlation coefficient in the first probability density distribution data, and calculate the mean correlation coefficient in the second probability density distribution data.
  • the first probability is the probability that the suspicious handwriting corresponding to the mean correlation coefficient and the sample handwriting are written by the same person
  • the second probability is the probability that the suspicious handwriting corresponding to the mean correlation coefficient and the sample handwriting are not written by the same person
  • the third data processing unit 73 connected to the second data processing unit 72, is used to calculate the ratio of the first probability to the second probability to obtain the likelihood ratio of the suspicious handwriting;
  • the result judging module 74 is connected to the third data processing unit 73, and is used for matching in a preset likelihood ratio scale according to the likelihood ratio to obtain the individual identification result of the suspicious handwriting.
  • the first probability density distribution data is the probability density distribution of the correlation coefficients between the handwritings of several real samples of the person written by the same person, and
  • the second probability density distribution data is the probability density distribution of the correlation coefficients between the handwriting of several other people and the handwriting of the real sample obtained by different people based on the handwriting of the real sample.
  • the handwriting of another person includes handwriting for signature, and/or copying handwriting, and/or copying handwriting.
  • the method of image preprocessing includes removing the background, and/or filling the whiteness in the strokes, and/or smoothing the strokes, and/or binarization.
  • the first dynamic feature includes width, and/or gray scale, and/or radian.
  • the second dynamic feature includes width, and/or gray scale, and/or radian.
  • An offline handwriting individual recognition method based on two-dimensional dynamic features which is applied to any of the above offline handwriting individual recognition system, as shown in Figure 3, specifically includes the following steps:
  • Step S1 the offline handwriting individual recognition system obtains a first handwriting image corresponding to a suspicious handwriting, and obtains a second handwriting image corresponding to a number of sample handwritings;
  • Step S2 the offline handwriting recognition system performs image preprocessing on the first handwriting image to obtain a first preprocessing image corresponding to the suspicious handwriting, and performs image preprocessing on each second handwriting image to obtain a second preprocessing corresponding to each sample handwriting image;
  • Step S3 the offline handwriting recognition system performs skeleton extraction on the first preprocessed image to obtain a first skeleton image corresponding to the suspicious handwriting, and performs skeleton extraction on each second preprocessed image to obtain a second skeleton image corresponding to the sample handwriting;
  • the line width of the first skeleton image and each second skeleton image is a single pixel
  • Step S4 the offline handwriting recognition system separately performs handwriting tracking according to the first skeleton image to obtain a first writing trajectory corresponding to the suspicious handwriting, and performs handwriting tracking according to each second skeleton image to obtain a second writing trajectory corresponding to each sample handwriting;
  • Step S5 the offline handwriting recognition system performs feature extraction on the corresponding first handwriting image according to the first writing trajectory to obtain the first dynamic feature of each pixel in the first writing trajectory, and according to each second writing trajectory, The second dynamic feature of each pixel in each second writing track is extracted from each corresponding second handwriting image;
  • Step S6 the offline handwriting recognition system processes the first dynamic feature corresponding to the suspicious handwriting and the second dynamic feature corresponding to each sample handwriting to obtain the correlation coefficient between the suspicious handwriting and each sample handwriting;
  • step S7 the offline handwriting individual recognition system processes and obtains the individual recognition result of the suspicious handwriting according to the first probability density distribution data and the second probability density distribution data obtained by statistics in advance, and the correlation coefficients.
  • step S7 specifically includes:
  • Step S71 the offline handwriting individual recognition system combines the first dynamic feature and each second dynamic feature in pairs to generate multiple feature pairs;
  • Step S72 the offline handwriting individual recognition system respectively performs dynamic time adjustment between the first dynamic feature and the second dynamic feature in each feature pair, so that the strokes corresponding to the first dynamic feature and the second dynamic feature and the corresponding strokes Match the parts one by one;
  • step S73 the offline handwriting individual recognition system separately calculates the correlation coefficient between the first dynamic feature and the second dynamic feature in each feature pair after dynamic time warping.
  • step S8 specifically includes:
  • Step S81 the offline handwriting individual recognition system averages all the correlation coefficients corresponding to the suspicious handwriting to obtain the mean correlation coefficient corresponding to the suspicious handwriting;
  • Step S82 the offline handwriting individual recognition system calculates the first probability corresponding to the mean correlation coefficient in the first probability density distribution data, and calculates the second probability corresponding to the mean correlation coefficient in the second probability density distribution data;
  • the first probability is the probability that the suspicious handwriting corresponding to the mean correlation coefficient and the sample handwriting are written by the same person
  • the second probability is the probability that the suspicious handwriting corresponding to the mean correlation coefficient and the sample handwriting are not written by the same person
  • Step S83 the offline handwriting individual recognition system calculates the ratio of the first probability to the second probability to obtain the likelihood ratio of the suspicious handwriting
  • step S84 the offline handwriting individual recognition system performs matching with a preset likelihood ratio scale according to the likelihood ratio to obtain an individual recognition result of the suspicious handwriting.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Human Computer Interaction (AREA)
  • Collating Specific Patterns (AREA)

Abstract

本发明提供一种基于二维动态特征的离线笔迹个体识别系统及方法，涉及身份识别技术领域，包括获取可疑笔迹对应的第一笔迹图像，若干样本笔迹对应的第二笔迹图像；将第一笔迹图像和第二笔迹图像预处理得到第一预处理图像和第二预处理图像；将第一预处理图像和第二预处理图像提取第一骨架图像和第二骨架图像；根据第一骨架图像和第二骨架图像得到第一书写轨迹和第二书写轨迹；根据第一书写轨迹和第二书写轨迹提取第一动态特征和第二动态特征；处理得到可疑笔迹与样本笔迹之间的相关系数，并根据相关系数处理得到个体识别结果。本发明包含的信息量更大，特征种类更丰富，有效提高识别准确率。

Description

一种基于二维动态特征的离线笔迹个体识别系统及方法 技术领域
本发明涉及基于行为特征的身份识别技术领域，尤其涉及一种基于二维动态特征的离线笔迹个体识别系统及方法。
背景技术
随着科学技术的进步，基于生物特征的身份识别技术也不断发展。每个个体都有唯一的可以测量或可自动识别、验证的生理特征或行为特征，其中生理特征包括眼虹膜、视网膜、指纹及面貌等，行为特征包括步态、声音及笔迹等。离线笔迹个体识别技术是根据人的书写技能习惯特性在纸上书写的自己、符号、绘画中的反映，通过存疑笔迹与样本笔迹的比较、鉴别，从而确定笔迹书写人的一项个体识别技术。笔迹是一个人独特的行为特征，不同人的笔迹存在很大差别，每个人书写习惯不同，一般情况下，笔迹模仿者只能模仿字形，却无法准确还原原作者的书写习惯，模仿的笔迹与原笔迹在细节上会存在差异，因此可以利用该部分的差异性和独特性，通过测量书写者的字形及笔画的速度、顺序和压力等特征进行身份识别。
根据笔迹获取途径的不同，现有笔迹识别技术分为在线笔迹识别和离线笔迹识别两种。其中，在线笔迹识别能够实时获取更多关于书写顺序、速度、压力、角度等有益于身份识别的特征信息，但这些信息需要运用特殊的设备来获取，因此限定了其应用范围及发展前景； 而离线笔迹则是以书写运动的轨迹，即二维静态图像的形式展现在人们面前，因此离线笔迹识别需要从中挖掘和提取特征信息。现有技术中，对离线笔迹识别存在准确率不高的问题。
发明内容
针对现有技术中存在的问题，本发明提供一种基于二维动态特征的离线笔迹个体识别系统，具体包括：
数据获取模块，用于获取可疑笔迹进对应的第一笔迹图像，以及获取若干样本笔迹对应的第二笔迹图像；
数据预处理模块，连接所述数据获取模块，用于将所述第一笔迹图像进行图像预处理得到所述可疑笔迹对应的第一预处理图像，以及将各所述第二笔迹图像分别进行所述图像预处理得到各所述样本笔迹对应的第二预处理图像；
骨架提取模块，连接所述数据预处理模块，用于将所述第一预处理图像进行骨架提取得到所述可疑笔迹对应的第一骨架图像，以及将各所述第二预处理图像分别进行所述骨架提取得到所述样本笔迹对应的第二骨架图像；
所述第一骨架图像和各所述第二骨架图像的线幅均为单个像素；
笔顺识别模块，连接所述骨架提取模块，用于根据所述第一骨架图像进行笔迹追踪得到所述可疑笔迹对应的第一书写轨迹，以及根据各所述第二骨架图像分别进行所述笔迹追踪得到各所述样本笔迹对应的第二书写轨迹；
特征提取模块，分别连接数据获取模块和所述笔顺识别模块，用于根据所述第一书写轨迹在对应的所述第一笔迹图像上进行特征提取，得到所述第一书写轨迹中每个像素点的第一动态特征，以及根据各所述第二书写轨迹分别在对应的各所述第二笔迹图像上提取得到各所述第二书写轨迹中每个像素点的第二动态特征；
第一数据处理模块，连接所述特征提取模块，用于根据所述可疑笔迹对应的所述第一动态特征，以及各所述样本笔迹对应的所述第二动态特征，处理得到所述可疑笔迹与各所述样本笔迹之间的相关系数；
第二数据处理模块，连接所述第一数据处理模块，用于根据预先统计得到的第一概率密度分布数据和第二概率密度分布数据，以及各所述相关系数，处理得到所述可疑笔迹的个体识别结果。
优选的，所述第一数据处理模块具体包括：
特征对生成单元，用于将所述第一动态特征与各所述第二动态特征之间进行两两组合生成多个特征对；
动态时间规整单元，连接所述特征对生成单元，用于将各所述特征对中的所述第一动态特征和所述第二动态特征之间进行动态时间规整，使得所述第一动态特征和所述第二动态特征对应的笔画及所述笔画对应的相应部位一一相匹配；
数据计算单元，连接所述动态时间规整单元，用于分别计算进行所述动态时间规整后的各所述特征对中的所述第一动态特征和所述第二动态特征之间的相关系数。
优选的，所述第二数据处理模块具体包括：
第一数据处理单元，用于将所述可疑笔迹对应的所有所述相关系数，进行均值处理得到所述可疑笔迹对应的均值相关系数；
第二数据处理单元，连接所述第一数据处理单元，用于计算所述均值相关系数在所述第一概率密度分布数据中对应的第一概率，以及计算所述均值相关系数在所述第二概率密度分布数据中对应的第二概率；
所述第一概率为所述均值相关系数对应的所述可疑笔迹与所述样本笔迹是同一人所写的概率，所述第二概率为所述均值相关系数对应的所述可疑笔迹与所述样本笔迹不是同一人所写的概率；
第三数据处理单元，连接所述第二数据处理单元，用于计算所述第一概率与所述第二概率的比值得到所述可疑笔迹的似然比；
结果判定模块，连接所述第三数据处理单元，用于根据所述似然比于预先设置的似然比量表中进行匹配，以得到所述可疑笔迹的个体识别结果。
优选的，所述第一概率密度分布数据为同一人书写得到的若干本人真实样本笔迹两两之间的相关系数的概率密度分布，以及
所述第二概率密度分布数据为不同人根据所述本人真实样本笔迹书写得到的若干他人笔迹与所述本人真实样本笔迹两两之间的相关系数的概率密度分布。
优选的，所述他人笔迹包括代签笔迹，和/或临摹笔迹，和/或套摹笔迹。
优选的，所述图像预处理的方式包括去除背景，和/或填补笔画中的露白，和/或平滑笔画，和/或二值化。
优选的，所述第一动态特征包括宽度，和/或灰度，和/或弧度。
优选的，所述第二动态特征包括宽度，和/或灰度，和/或弧度。
一种基于二维动态特征的离线笔迹个体识别方法，应用于以上任意一项所述的离线笔迹个体识别系统，具体包括以下步骤：
步骤S1，所述离线笔迹个体识别系统获取可疑笔迹对应的第一笔迹图像，以及获取若干样本笔迹对应的第二笔迹图像；
步骤S2，所述离线笔迹个体识别系统将所述第一笔迹图像进行图像预处理得到所述可疑笔迹对应的第一预处理图像，以及将各所述第二笔迹图像分别进行所述图像预处理得到各所述样本笔迹对应的第二预处理图像；
步骤S3，所述离线笔迹个体识别系统将所述第一预处理图像进行骨架提取得到所述可疑笔迹对应的第一骨架图像，以及将各所述第二预处理图像分别进行所述骨架提取得到所述样本笔迹对应的第二骨架图像；
第一骨架图像和各第二骨架图像的线幅均为单个像素；
步骤S4，所述离线笔迹个体识别系统根据所述第一骨架图像分别进行笔迹追踪得到所述可疑笔迹对应的第一书写轨迹，以及根据各所述第二骨架图像分别进行所述笔迹追踪得到各所述样本笔迹对应的第二书写轨迹；
步骤S5，所述离线笔迹个体识别系统根据所述第一书写轨迹在 对应的所述第一笔迹图像上进行特征提取，得到所述第一书写轨迹中每个像素点的第一动态特征，以及根据各所述第二书写轨迹分别在对应的各所述第二笔迹图像上提取得到各所述第二书写轨迹中每个像素点的第二动态特征；
步骤S6，所述离线笔迹个体识别系统根据所述可疑笔迹对应的所述第一动态特征，以及各所述样本笔迹对应的所述第二动态特征，处理得到所述可疑笔迹与各所述样本笔迹之间的相关系数；
步骤S7，所述离线笔迹个体识别系统根据预先统计得到的第一概率密度分布数据和第二概率密度分布数据，以及各所述相关系数，处理得到所述可疑笔迹的个体识别结果。
优选的，所述步骤S7具体包括：
步骤S71，所述离线笔迹个体识别系统将所述第一动态特征与各所述第二动态特征之间进行两两组合生成多个特征对；
步骤S72，所述离线笔迹个体识别系统分别将各所述特征对中的所述第一动态特征与所述第二动态特征之间进行动态时间规整，使得所述第一动态特征与所述第二动态特征对应的笔画及所述笔画的相应部位一一相匹配；
步骤S73，所述离线笔迹个体识别系统分别计算进行所述动态时间规整后的各所述特征对中的所述第一动态特征与所述第二动态特征之间的相关系数。
优选的，所述步骤S8具体包括：
步骤S81，所述离线笔迹个体识别系统将所述可疑笔迹对应的所 有所述相关系数进行均值处理得到所述可疑笔迹对应的均值相关系数；
步骤S82，所述离线笔迹个体识别系统计算所述均值相关系数在所述第一概率密度分布数据中对应的第一概率，以及计算所述均值相关系数在所述第二概率密度分布数据中对应的第二概率；
所述第一概率为所述均值相关系数对应的所述可疑笔迹与所述样本笔迹是同一人所写的概率，所述第二概率为所述均值相关系数对应的所述可疑笔迹与所述样本笔迹不是同一人所写的概率；
步骤S83，所述离线笔迹个体识别系统计算所述第一概率与所述第二概率的比值得到所述可疑笔迹的似然比；
步骤S84，所述离线笔迹个体识别系统根据所述似然比于预先设置的似然比量表中进行匹配，以得到所述可疑笔迹的个体识别结果。
上述技术方案具有如下优点或有益效果：
1)只需通过二维扫描设备即可获取可疑笔迹与样本笔迹的笔迹图像，且图像预处理技术有效去除纸张背景等因素的干扰，适应不同粗细的书写工具的书写结果，增强了对不同检测对象的兼容性；
2)特征的自动提取，最大限度的排除了人为主观因素的参与度，从源头上避免了人为因素的干扰，有效提高识别准确率；
3)通过预先统计得到的概率密度分布对可疑笔迹与样本笔迹之间的相关系数做进一步处理，进一步提高了识别准确率。
附图说明
图1为本发明的较佳的实施例中，一种基于二维动态特征的离线笔迹个体识别系统的结构示意图；
图2为本发明的较佳的实施例中，似然比量表的示意图；
图3为本发明的较佳的实施例中，一种基于二维动态特征的离线笔迹个体识别方法的流程示意图；
图4为本发明的较佳的实施例中，一种基于二维动态特征的离线笔迹个体识别方法的子流程示意图；
图5为本发明的较佳的实施例中，一种基于二维动态特征的离线笔迹个体识别方法的子流程示意图。
具体实施方式
下面结合附图和具体实施例对本发明进行详细说明。本发明并不限定于该实施方式，只要符合本发明的主旨，则其他实施方式也可以属于本发明的范畴。
本发明的较佳的实施例中，基于现有技术中存在的上述问题，现提供一种基于二维动态特征的离线笔迹个体识别系统，如图1所示，具体包括：
数据获取模块1，用于获取可疑笔迹对应的第一笔迹图像，以及获取若干样本笔迹对应的第二笔迹图像；
数据预处理模块2，连接数据获取模块1，用于将第一笔迹图像进行图像预处理得到可疑笔迹对应的第一预处理图像，以及将各第二笔迹图像分别进行图像预处理得到各样本笔迹对应的第二预处理图 像；
骨架提取模块3，连接数据预处理模块2，用于将第一预处理图像进行骨架提取得到可疑笔迹对应的第一骨架图像，以及将各第二预处理图像分别进行骨架提取得到样本笔迹对应的第二骨架图像；
第一骨架图像和各第二骨架图像的线幅均为单个像素；
笔顺识别模块4，连接骨架提取模块3，用于根据第一骨架图像进行笔迹追踪得到可疑笔迹对应的第一书写轨迹，以及根据各第二骨架图像分别进行笔迹追踪得到各样本笔迹对应的第二书写轨迹；
特征提取模块5，分别连接数据获取模块1和笔顺识别模块4，用于根据第一书写轨迹在对应的第一笔迹图像上进行特征提取，得到第一书写轨迹中每个像素点的第一动态特征，以及根据各第二书写轨迹分别在对应的各第二笔迹图像上提取得到各第二书写轨迹中每个像素点的第二动态特征；
第一数据处理模块6，连接特征提取模块5，用于根据可疑笔迹对应的第一动态特征，以及各样本笔迹对应的第二动态特征，处理得到可疑笔迹与各样本笔迹之间的相关系数；
第二数据处理模块7，连接第一数据处理模块6，用于根据预先统计得到的第一概率密度分布数据和第二概率密度分布数据，以及各相关系数，处理得到可疑笔迹的个体识别结果。
具体地，本实施例中，优选为首先通过采用大面积光学二维形貌扫描仪器进行扫描操作，具体为依次将各待检测的可疑笔迹和样本笔迹置于上述大面积光学二维形貌扫描仪器的载物台上，在扫描完成 后，与上述大面积光学二维形貌扫描仪器连接的计算机获取可疑笔迹对应的第一笔迹图像，以及获取每个样本笔迹对应的第二笔迹图像。随后将第一笔迹图像和第二笔迹图像分别进行图像预处理得到对应于第一笔迹图像的第一预处理图像，以及对应于第二笔迹图像的第二预处理图像，上述图像预处理包括去除背景、填补笔画中的露白、平滑笔画和二值化等操作。本实施例中，也可以通过相机拍摄待检测的可以笔迹和样本笔迹，进而获取可疑笔迹对应的第一笔迹图像，以及获取每个样本笔迹对应的第二笔迹图像。
本实施例中，对于可疑笔迹和样本笔迹的处理过程类似，因此此处仅以可疑笔迹的处理过程为例进行详细描述，对于样本笔迹的处理过程此处不再赘述。可疑笔迹的处理过程如下：
在第一预处理图像上提取对应的可疑笔迹的骨架，以得到对应于可疑笔迹的第一骨架图像，通过在第一骨架图像上进行笔迹追踪即得到对应于可疑笔迹的第一书写轨迹。上述笔迹追踪通过人机交互的形式完成，具体包括：
首先使用鼠标、电子笔或触摸屏的方式在第一骨架图像上的笔画起始位置附近点击，以自动获取笔画起点，随后沿笔画运行方向进行追踪；若当前追踪笔画有误或由于人为导致误操作，可以通过撤销当前追踪笔画，以返回上一个笔画并继续进行笔迹追踪，直至完成对整个第一骨架图像上的每个笔画的笔迹追踪，最终获得上述第一书写轨迹。
进一步地，根据获取的第一书写轨迹在对应的第一笔迹图上获取 第一书写轨迹上每个像素点的第一动态特征，即笔画宽度、笔画灰度和笔画弧度等特征，上述第一动态特征作为可疑笔迹的二维动态特征。
本实施例中，按照上述相同的处理过程，对样本笔迹进行处理得到样本笔迹的二维动态特征。随后计算可疑笔迹的二维动态特征与样本笔迹的二维动态特征之间的相关系数，通过相关系数表征可疑笔迹与样本笔迹的相似度。具体地，由于笔迹的形成是一个过程，即使是同一人书写的笔迹，由于每次的书写速度不完全相同，各个对应笔画的长度不同，造成笔迹对应的序列长度不同，相同笔画在序列上发生错位。从而在以序列长度为横轴的坐标系上，可疑笔迹的二维动态特征投射在上述坐标系上时每个笔画的序列长度，与样本笔迹的二维动态特征投射在上述坐标系上时每个笔画的序列长度是不完全相同的，造成无法计算出有效的可疑笔迹的二维动态特征与样本笔迹的二维动态特征之间的相关系数。因此，在计算相关系数之前，需要对需要进行相关系数计算的一组可疑笔迹的二维动态特征和样本笔迹的二维动态特征进行动态时间规整，本实施例中，采用DTW技术进行动态时间规整，使得可疑笔迹的二维动态特征和样本笔迹的二维动态特征在对应的笔画及该笔画的相应部位一一相匹配，且相同笔画在序列长度上相同。
本实施例中，在获取可疑笔迹与样本笔迹的相似度后，为进一步提高识别准确率，还需要将上述相似度，即相关系数代入预先统计得到的概率密度分布中进行进一步处理。在进行笔迹识别时，仅获取一 个可疑笔迹和一个样本笔迹时，对比的结果仅有一个相关系数，该相关系数即为本发明所述的均值相关系数。需要说明的是，通常进行笔迹识别时，会获取到若干可疑笔迹和若干样本笔迹，如获取两个可疑笔迹和三个样本笔迹，通过将两个可疑笔迹的二维动态特征分别与三个样本笔迹的二维动态特征进行相关系数的计算，对于每个可疑笔迹，将得到对应的三个相关系数的计算结果。
本实施例中，对于每个可疑笔迹，无需将上述计算结果中的三个相关系数均代入预先统计得到的概率密度分布中进行进一步处理，而是先将三个相关系数取均值后得到相应的均值相关系数，再代入预先统计得到的概率密度分布中进行进一步处理。
具体地，上述预先统计得到的第一概率密度分布数据为同一人书写得到的若干本人真实样本笔迹两两之间的相关系数的概率密度分布，以及第二概率密度分布数据为不同人根据上述本人真实样本笔迹书写得到的若干他人笔迹与上述本人真实样本笔迹两两之间的相关系数的概率密度分布。本实施例中，预先获取数十万数量的笔迹数据组建相应的笔迹数据库，该笔迹数据库中包括同一人书写得到的若干本人真实样本笔迹，即本人笔迹，还包括不同人根据本人真实样本笔迹书写得到的若干他人笔迹，即根据上述本人笔迹得到的代签笔迹，和/或临摹笔迹，和/或套摹笔迹等他人笔迹。通过分别计算两两本人笔迹之间的相关系数，并根据得到的若干相关系数得到本人笔迹的概率密度分布；通过分别计算每个他人笔迹与本人笔迹两两之间的相关系数，并根据得到的若干相关系数得到他人笔迹的概率密度分布。随 后通过将上述得到的每个可疑笔迹的均值相关系数分别代入到本人笔迹的概率密度分布和他人笔迹的概率密度分布中，以得到表征该可疑笔迹与样本笔迹是同一人所写的第一概率，以及表征该可疑笔迹与样本笔迹不是同一人所写的第二概率。
本实施例中，进一步通过下述公式计算可疑笔迹的似然比：
Figure PCTCN2019122178-appb-000001
其中，SLR表示可疑笔迹的似然比；H p表示可疑笔迹与样本笔迹是同一人所写；H p表示可疑笔迹与样本笔迹不是同一人所写；E U表示可疑笔迹；E S表示样本笔迹；
本实施例中，如图2所示，预先设有似然比量表，将上述似然比的计算结果与该似然比量表进行匹配，若似然比大于10000，则给出的个体识别结果为极强支持该可疑笔迹与样本笔迹是同一人所写；若似然比小于0.001，则给出的个体识别结果为极强支持该可疑笔迹与样本笔迹不是同一人所写；其他匹配结果以此类推。
本发明的较佳的实施例中，第一数据处理模块6具体包括：
特征对生成单元61，用于将第一动态特征与各第二动态特征之间进行两两组合生成多个特征对；
动态时间规整单元62，连接特征对生成单元61，用于将各特征对中的第一动态特征和第二动态特征之间进行动态时间规整，使得第一动态特征和第二动态特征对应的笔画及笔画的相应部位一一相匹配；
数据计算单元63，连接动态时间规整单元62，用于分别计算进 行动态时间规整后的各特征对中的第一动态特征和第二动态特征之间的相关系数。
本发明的较佳的实施例中，第二数据处理模块7具体包括：
第一数据处理单元71，用于将可疑笔迹对应的所有相关系数，进行均值处理得到可疑笔迹对应的均值相关系数；
第二数据处理单元72，连接第一数据处理单元71，用于计算均值相关系数在第一概率密度分布数据中对应的第一概率，以及计算均值相关系数在第二概率密度分布数据中对应的第二概率；
第一概率为均值相关系数对应的可疑笔迹与样本笔迹是同一人所写的概率，第二概率为均值相关系数对应的可疑笔迹与样本笔迹不是同一人所写的概率；
第三数据处理单元73，连接第二数据处理单元72，用于计算第一概率与第二概率的比值得到可疑笔迹的似然比；
结果判定模块74，连接第三数据处理单元73，用于根据似然比于预先设置的似然比量表中进行匹配，以得到可疑笔迹的个体识别结果。
本发明的较佳的实施例中，第一概率密度分布数据为同一人书写得到的若干本人真实样本笔迹两两之间的相关系数的概率密度分布，以及
第二概率密度分布数据为不同人根据本人真实样本笔迹书写得到的若干他人笔迹与本人真实样本笔迹两两之间的相关系数的概率密度分布。
本发明的较佳的实施例中，他人笔迹包括代签笔迹，和/或临摹笔迹，和/或套摹笔迹。
本发明的较佳的实施例中，图像预处理的方式包括去除背景，和/或填补笔画中的露白，和/或平滑笔画，和/或二值化。
本发明的较佳的实施例中，第一动态特征包括宽度，和/或灰度，和/或弧度。
本发明的较佳的实施例中，第二动态特征包括宽度，和/或灰度，和/或弧度。
一种基于二维动态特征的离线笔迹个体识别方法，应用于以上任意一项的离线笔迹个体识别系统，如图3所示，具体包括以下步骤：
步骤S1，离线笔迹个体识别系统获取可疑笔迹对应的第一笔迹图像，以及获取若干样本笔迹对应的第二笔迹图像；
步骤S2，离线笔迹个体识别系统将第一笔迹图像进行图像预处理得到可疑笔迹对应的第一预处理图像，以及将各第二笔迹图像分别进行图像预处理得到各样本笔迹对应的第二预处理图像；
步骤S3，离线笔迹个体识别系统将第一预处理图像进行骨架提取得到可疑笔迹对应的第一骨架图像，以及将各第二预处理图像分别进行骨架提取得到样本笔迹对应的第二骨架图像；
第一骨架图像和各第二骨架图像的线幅均为单个像素；
步骤S4，离线笔迹个体识别系统根据第一骨架图像分别进行笔迹追踪得到可疑笔迹对应的第一书写轨迹，以及根据各第二骨架图像分别进行笔迹追踪得到各样本笔迹对应的第二书写轨迹；
步骤S5，离线笔迹个体识别系统根据第一书写轨迹在对应的第一笔迹图像上进行特征提取，得到第一书写轨迹中每个像素点的第一动态特征，以及根据各第二书写轨迹分别在对应的各第二笔迹图像上提取得到各第二书写轨迹中每个像素点的第二动态特征；
步骤S6，离线笔迹个体识别系统根据可疑笔迹对应的第一动态特征，以及各样本笔迹对应的第二动态特征，处理得到可疑笔迹与各样本笔迹之间的相关系数；
步骤S7，离线笔迹个体识别系统根据预先统计得到的第一概率密度分布数据和第二概率密度分布数据，以及各相关系数，处理得到可疑笔迹的个体识别结果。
本发明的较佳的实施例中，如图4所示，步骤S7具体包括：
步骤S71，离线笔迹个体识别系统将第一动态特征与各第二动态特征之间进行两两组合生成多个特征对；
步骤S72，离线笔迹个体识别系统分别将各特征对中的第一动态特征与第二动态特征之间进行动态时间规整，使得第一动态特征与第二动态特征对应的笔画及所述笔画的相应部位一一相匹配；
步骤S73，离线笔迹个体识别系统分别计算进行动态时间规整后的各特征对中的第一动态特征与第二动态特征之间的相关系数。
本发明的较佳的实施例中，如图5所示，步骤S8具体包括：
步骤S81，离线笔迹个体识别系统将可疑笔迹对应的所有相关系数进行均值处理得到可疑笔迹对应的均值相关系数；
步骤S82，离线笔迹个体识别系统计算均值相关系数在第一概率 密度分布数据中对应的第一概率，以及计算均值相关系数在第二概率密度分布数据中对应的第二概率；
第一概率为均值相关系数对应的可疑笔迹与样本笔迹是同一人所写的概率，第二概率为均值相关系数对应的可疑笔迹与样本笔迹不是同一人所写的概率；
步骤S83，离线笔迹个体识别系统计算第一概率与第二概率的比值得到可疑笔迹的似然比；
步骤S84，离线笔迹个体识别系统根据似然比于预先设置的似然比量表中进行匹配，以得到可疑笔迹的个体识别结果。
以上所述仅为本发明较佳的实施例，并非因此限制本发明的实施方式及保护范围，对于本领域技术人员而言，应当能够意识到凡运用本说明书及图示内容所作出的等同替换和显而易见的变化所得到的方案，均应当包含在本发明的保护范围内。

Claims (11)

  1. 一种基于二维动态特征的离线笔迹个体识别系统，其特征在于，具体包括：
    数据获取模块，用于获取可疑笔迹对应的第一笔迹图像，以及获取若干样本笔迹对应的第二笔迹图像；
    数据预处理模块，连接所述数据获取模块，用于将所述第一笔迹图像进行图像预处理得到所述可疑笔迹对应的第一预处理图像，以及将各所述第二笔迹图像分别进行所述图像预处理得到各所述样本笔迹对应的第二预处理图像；
    骨架提取模块，连接所述数据预处理模块，用于将所述第一预处理图像进行骨架提取得到所述可疑笔迹对应的第一骨架图像，以及将各所述第二预处理图像分别进行所述骨架提取得到所述样本笔迹对应的第二骨架图像；
    所述第一骨架图像和各所述第二骨架图像的线幅均为单个像素；
    笔顺识别模块，连接所述骨架提取模块，用于根据所述第一骨架图像进行笔迹追踪得到所述可疑笔迹对应的第一书写轨迹，以及根据各所述第二骨架图像分别进行所述笔迹追踪得到各所述样本笔迹对应的第二书写轨迹；
    特征提取模块，分别连接数据获取模块和所述笔顺识别模块，用于根据所述第一书写轨迹在对应的所述第一笔迹图像上进行特征提取，得到所述第一书写轨迹中每个像素点的第一动态特征，以及根据各所述第二书写轨迹分别在对应的各所述第二笔迹图像上提取得到各所述第二书写轨迹中每个像素点的第二动态特征；
    第一数据处理模块，连接所述特征提取模块，用于根据所述可疑笔迹对应的所述第一动态特征，以及各所述样本笔迹对应的所述第二动态特征，处理得到所述可疑笔迹与各所述样本笔迹之间的相关系数；
    第二数据处理模块，连接所述第一数据处理模块，用于根据预先统计得到的第一概率密度分布数据和第二概率密度分布数据，以及各所述相关系数，处理得到所述可疑笔迹的个体识别结果。
  2. 根据权利要求1所述的离线笔迹个体识别系统，其特征在于，所述第一数据处理模块具体包括：
    特征对生成单元，用于将所述第一动态特征与各所述第二动态特征之间进行两两组合生成多个特征对；
    动态时间规整单元，连接所述特征对生成单元，用于将各所述特征对中的所述第一动态特征和所述第二动态特征之间进行动态时间规整，使得所述第一动态特征和所述第二动态特征对应的笔画及所述笔画的相应部位一一相匹配；
    数据计算单元，连接所述动态时间规整单元，用于分别计算进行所述动态时间规整后的各所述特征对中的所述第一动态特征和所述第二动态特征之间的相关系数。
  3. 根据权利要求1所述的离线笔迹个体识别系统，其特征在于，所述第二数据处理模块具体包括：
    第一数据处理单元，用于将所述可疑笔迹对应的所有所述相关系数，进行均值处理得到所述可疑笔迹对应的均值相关系数；
    第二数据处理单元，连接所述第一数据处理单元，用于计算所述均值相关系数在所述第一概率密度分布数据中对应的第一概率，以及计算所述均值相关系数在所述第二概率密度分布数据中对应的第二概率；
    所述第一概率为所述均值相关系数对应的所述可疑笔迹与所述样本笔迹是同一人所写的概率，所述第二概率为所述均值相关系数对应的所述可疑笔迹与所述样本笔迹不是同一人所写的概率；
    第三数据处理单元，连接所述第二数据处理单元，用于计算所述第一概率与所述第二概率的比值得到所述可疑笔迹的似然比；
    结果判定模块，连接所述第三数据处理单元，用于根据所述似然比于预先设置的似然比量表中进行匹配，以得到所述可疑笔迹的个体识别结果。
  4. 根据权利要求3所述的离线笔迹个体识别系统，其特征在于，所述第一概率密度分布数据为同一人书写得到的若干本人真实样本笔迹两两之间的相关系数的概率密度分布，以及
    所述第二概率密度分布数据为不同人根据所述本人真实样本笔迹书写得到的若干他人笔迹与所述本人真实样本笔迹两两之间的相关系数的概率密度分布。
  5. 根据权利要求4所述的离线笔迹个体识别系统，其特征在于，所述他人笔迹包括代签笔迹，和/或临摹笔迹，和/或套摹笔迹。
  6. 根据权利要求1所述的离线笔迹个体识别系统，其特征在于，所述图像预处理的方式包括去除背景，和/或填补笔画中的露白，和/ 或平滑笔画，和/或二值化。
  7. 根据权利要求1所述的离线笔迹个体识别系统，其特征在于，所述第一动态特征包括宽度，和/或灰度，和/或弧度。
  8. 根据权利要求1所述的离线笔迹个体识别系统，其特征在于，所述第二动态特征包括宽度，和/或灰度，和/或弧度。
  9. 一种基于二维动态特征的离线笔迹个体识别方法，其特征在于，应用于如权利要求1-8中任意一项所述的离线笔迹个体识别系统，具体包括以下步骤：
    步骤S1，所述离线笔迹个体识别系统获取可疑笔迹对应的第一笔迹图像，以及获取若干样本笔迹对应的第二笔迹图像；
    步骤S2，所述离线笔迹个体识别系统将所述第一笔迹图像进行图像预处理得到所述可疑笔迹对应的第一预处理图像，以及将各所述第二笔迹图像分别进行所述图像预处理得到各所述样本笔迹对应的第二预处理图像；
    步骤S3，所述离线笔迹个体识别系统将所述第一预处理图像进行骨架提取得到所述可疑笔迹对应的第一骨架图像，以及将各所述第二预处理图像分别进行所述骨架提取得到所述样本笔迹对应的第二骨架图像；
    第一骨架图像和各第二骨架图像的线幅均为单个像素；
    步骤S4，所述离线笔迹个体识别系统根据所述第一骨架图像分别进行笔迹追踪得到所述可疑笔迹对应的第一书写轨迹，以及根据各所述第二骨架图像分别进行所述笔迹追踪得到各所述样本笔迹对应 的第二书写轨迹；
    步骤S5，所述离线笔迹个体识别系统根据所述第一书写轨迹在对应的所述第一笔迹图像上进行特征提取，得到所述第一书写轨迹中每个像素点的第一动态特征，以及根据各所述第二书写轨迹分别在对应的各所述第二笔迹图像上提取得到各所述第二书写轨迹中每个像素点的第二动态特征；
    步骤S6，所述离线笔迹个体识别系统根据所述可疑笔迹对应的所述第一动态特征，以及各所述样本笔迹对应的所述第二动态特征，处理得到所述可疑笔迹与各所述样本笔迹之间的相关系数；
    步骤S7，所述离线笔迹个体识别系统根据预先统计得到的第一概率密度分布数据和第二概率密度分布数据，以及各所述相关系数，处理得到所述可疑笔迹的个体识别结果。
  10. 根据权利要求9所述的离线笔迹个体识别方法，其特征在于，所述步骤S7具体包括：
    步骤S71，所述离线笔迹个体识别系统将所述第一动态特征与各所述第二动态特征之间进行两两组合生成多个特征对；
    步骤S72，所述离线笔迹个体识别系统分别将各所述特征对中的所述第一动态特征与所述第二动态特征之间进行动态时间规整，使得所述第一动态特征与所述第二动态特征对应的笔画及所述笔画的相应部位一一相匹配；
    步骤S73，所述离线笔迹个体识别系统分别计算进行所述动态时间规整后的各所述特征对中的所述第一动态特征与所述第二动态特 征之间的相关系数。
  11. 根据权利要求9所述的离线笔迹个体识别方法，其特征在于，所述步骤S8具体包括：
    步骤S81，所述离线笔迹个体识别系统将所述可疑笔迹对应的所有所述相关系数进行均值处理得到所述可疑笔迹对应的均值相关系数；
    步骤S82，所述离线笔迹个体识别系统计算所述均值相关系数在所述第一概率密度分布数据中对应的第一概率，以及计算所述均值相关系数在所述第二概率密度分布数据中对应的第二概率；
    所述第一概率为所述均值相关系数对应的所述可疑笔迹与所述样本笔迹是同一人所写的概率，所述第二概率为所述均值相关系数对应的所述可疑笔迹与所述样本笔迹不是同一人所写的概率；
    步骤S83，所述离线笔迹个体识别系统计算所述第一概率与所述第二概率的比值得到所述可疑笔迹的似然比；
    步骤S84，所述离线笔迹个体识别系统根据所述似然比于预先设置的似然比量表中进行匹配，以得到所述可疑笔迹的个体识别结果。
PCT/CN2019/122178 2019-08-22 2019-11-29 一种基于二维动态特征的离线笔迹个体识别系统及方法 WO2021031446A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/969,492 US11854284B2 (en) 2019-08-22 2019-08-22 Offline handwriting individual recognition system and method based on two-dimensional dynamic features

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910780111.1 2019-08-22
CN201910780111.1A CN110458145B (zh) 2019-08-22 2019-08-22 一种基于二维动态特征的离线笔迹个体识别系统及方法

Publications (1)

Publication Number Publication Date
WO2021031446A1 true WO2021031446A1 (zh) 2021-02-25

Family

ID=68488532

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/122178 WO2021031446A1 (zh) 2019-08-22 2019-11-29 一种基于二维动态特征的离线笔迹个体识别系统及方法

Country Status (3)

Country Link
US (1) US11854284B2 (zh)
CN (1) CN110458145B (zh)
WO (1) WO2021031446A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115909372A (zh) * 2023-01-09 2023-04-04 浙江环玛信息科技有限公司 行为检测方法、系统、终端及存储介质

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11854284B2 (en) * 2019-08-22 2023-12-26 Academy Of Forensic Science Offline handwriting individual recognition system and method based on two-dimensional dynamic features
CN113468987B (zh) * 2021-06-17 2023-04-18 重庆傲雄在线信息技术有限公司 一种电子笔迹鉴定方法、系统、电子设备及存储介质
CN113407921B (zh) * 2021-06-30 2023-04-07 重庆建安仪器有限责任公司 一种笔迹识别的登录方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104392229A (zh) * 2014-10-09 2015-03-04 南通大学 基于笔画片段方向分布特征的笔迹识别方法
CN107256389A (zh) * 2017-05-26 2017-10-17 山西农业大学 基于笔迹特征识别的认证及加密方法
CN109271909A (zh) * 2018-09-03 2019-01-25 广东小天才科技有限公司 笔迹识别方法、装置、终端与存储介质
CN110458145A (zh) * 2019-08-22 2019-11-15 司法鉴定科学研究院 一种基于二维动态特征的离线笔迹个体识别系统及方法

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9489574B2 (en) * 2011-12-06 2016-11-08 Kyungpook National University Industry-Academic Cooperation Foundation Apparatus and method for enhancing user recognition
CN102592152B (zh) * 2012-01-05 2013-06-19 中国科学院合肥物质科学研究院 一种基于计算机系统的在线笔迹认证方法
ITRM20130022A1 (it) * 2013-01-11 2014-07-12 Natural Intelligent Technologies S R L Procedimento e apparato di riconoscimento di scrittura a mano
US9898187B2 (en) * 2013-06-09 2018-02-20 Apple Inc. Managing real-time handwriting recognition
CN103616955A (zh) * 2013-12-10 2014-03-05 步步高教育电子有限公司 一种笔迹或手势识别方法和识别设备
US10354133B2 (en) * 2015-08-26 2019-07-16 Beijing Lejent Technology Co., Ltd. Method for structural analysis and recognition of handwritten mathematical formula in natural scene image
US10169670B2 (en) * 2015-11-30 2019-01-01 International Business Machines Corporation Stroke extraction in free space
US10248880B1 (en) * 2016-06-06 2019-04-02 Boston Inventions, LLC Method of processing and recognizing hand-written characters
CN108804989B (zh) * 2017-05-05 2021-11-30 京东方科技集团股份有限公司 书画装置、书画设备及书画辅助方法
US10748031B1 (en) * 2019-05-22 2020-08-18 Read-Ink Corporation Handwriting recognition systems and methods

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104392229A (zh) * 2014-10-09 2015-03-04 南通大学 基于笔画片段方向分布特征的笔迹识别方法
CN107256389A (zh) * 2017-05-26 2017-10-17 山西农业大学 基于笔迹特征识别的认证及加密方法
CN109271909A (zh) * 2018-09-03 2019-01-25 广东小天才科技有限公司 笔迹识别方法、装置、终端与存储介质
CN110458145A (zh) * 2019-08-22 2019-11-15 司法鉴定科学研究院 一种基于二维动态特征的离线笔迹个体识别系统及方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
CHEN XIAO-HONG, JIA YU-WEN, YANG XU, XU CHE, CUI LAN, BIAN XIN-WEI, LUO YI-WEN: "Extraction and Analysis of dynamic features of Static Signature on Paper", CHINESE JOURNAL OF FORENSIC SCIENCES, vol. 4, 1 January 2012 (2012-01-01), pages 28 - 36, XP055783159 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115909372A (zh) * 2023-01-09 2023-04-04 浙江环玛信息科技有限公司 行为检测方法、系统、终端及存储介质

Also Published As

Publication number Publication date
CN110458145B (zh) 2022-12-27
US11854284B2 (en) 2023-12-26
US20230177857A1 (en) 2023-06-08
CN110458145A (zh) 2019-11-15

Similar Documents

Publication Publication Date Title
WO2021031446A1 (zh) 一种基于二维动态特征的离线笔迹个体识别系统及方法
Ross et al. Fingerprint warping using ridge curve correspondences
CN108230383B (zh) 手部三维数据确定方法、装置及电子设备
Feng et al. Robust and efficient algorithms for separating latent overlapped fingerprints
Alyuz et al. 3-D face recognition under occlusion using masked projection
WO2021031445A1 (zh) 一种基于三维动态特征的离线笔迹个体识别系统及方法
Rahman et al. Person identification using ear biometrics
Vretos et al. 3D facial expression recognition using Zernike moments on depth images
CN104392229B (zh) 基于笔画片段方向分布特征的笔迹识别方法
Chansri et al. Hand gesture recognition for Thai sign language in complex background using fusion of depth and color video
CN110796101A (zh) 一种嵌入式平台的人脸识别方法及系统
CN109376717A (zh) 人脸对比的身份识别方法、装置、电子设备及存储介质
Caya et al. Recognition of tongue print biometrie using binary robust independent elementary features
Rahman et al. Human ear recognition using geometric features
CN108090476A (zh) 一种针对具有外部遮挡的3d人脸识别方法
Bharadi et al. Multi-modal biometric recognition using human iris and dynamic pressure variation of handwritten signatures
Latha et al. A novel method for person authentication using retinal images
Ravidas et al. Deep learning for pose-invariant face detection in unconstrained environment
CN111428670B (zh) 人脸检测方法、装置、存储介质及设备
Kaur et al. Offline signature verification in Punjabi based on SURF features and critical point matching using HMM
Ismail et al. Signature Recognition using Multi Scale Fourier Descriptor And Wavelet Transform
JP7347750B2 (ja) 照合装置、学習装置、方法、及びプログラム
Lin et al. 3D face authentication by mutual coupled 3D and 2D feature extraction
Adhan et al. Alphabetic hand sign interpretation using geometric invariance
JP2019121187A (ja) 画像処理装置、画像処理方法および画像処理プログラム

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: 19942080

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19942080

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19942080

Country of ref document: EP

Kind code of ref document: A1