WO2016045215A1 - 指纹图像质量的判断方法和装置 - Google Patents

指纹图像质量的判断方法和装置 Download PDF

Info

Publication number
WO2016045215A1
WO2016045215A1 PCT/CN2014/094567 CN2014094567W WO2016045215A1 WO 2016045215 A1 WO2016045215 A1 WO 2016045215A1 CN 2014094567 W CN2014094567 W CN 2014094567W WO 2016045215 A1 WO2016045215 A1 WO 2016045215A1
Authority
WO
WIPO (PCT)
Prior art keywords
fingerprint image
quality
gradient direction
hog feature
optimal classification
Prior art date
Application number
PCT/CN2014/094567
Other languages
English (en)
French (fr)
Inventor
姜洪霖
王兵
Original Assignee
深圳市汇顶科技股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳市汇顶科技股份有限公司 filed Critical 深圳市汇顶科技股份有限公司
Publication of WO2016045215A1 publication Critical patent/WO2016045215A1/zh

Links

Images

Classifications

    • 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
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/12—Fingerprints or palmprints
    • G06V40/1365—Matching; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines

Definitions

  • the present invention relates to the field of communications technologies, and in particular, to a method and apparatus for judging fingerprint image quality.
  • Fingerprint recognition technology has begun to be widely used in mobile terminals.
  • the recognition algorithm is the core technology. Due to the limitation of the fingerprint acquisition sensor, when the finger is sweaty or muddy, the recognition rate will be greatly reduced. Especially for the small-sized fingerprint collection device of the mobile terminal, the recognition algorithm relies on the input fingerprint image quality, and the quality is more The higher the recognition rate, the lower the false positive rate, so the judgment of the fingerprint image quality before the recognition is particularly important.
  • the traditional method of judging firstly counts the mean values, variance, information entropy, and ring spectrum structure of the fingerprint image, and then integrates these indicators to calculate the quality score.
  • this type of method is only applicable to the full-capture fingerprint recognition system, but not to the small-size fingerprint acquisition system; further, these feature indicators do not fully consider that the fingerprint is an image with a specific texture structure. Therefore, the method for judging the quality of the fingerprint image in the prior art cannot accurately determine the quality of the fingerprint image in the small-sized fingerprint collection device (or system).
  • the main object of the present invention is to provide a method and a device for judging the quality of a fingerprint image, aiming at automatically determining the quality of the fingerprint image, and laying a good foundation for improving the fingerprint recognition rate and reducing the false recognition rate.
  • the present invention provides a method for judging the quality of a fingerprint image, comprising the steps of:
  • the SVM support vector machine classifier is used to learn according to the fingerprint image sample to obtain an optimal classification surface
  • the using the SVM support vector machine classifier to learn according to the fingerprint image sample to obtain an optimal classification plane comprises:
  • the HOG feature of the fingerprint image sample is input into the SVM classifier for training learning to obtain an optimal classification surface.
  • the calculating the HOG gradient direction histogram feature of the fingerprint image comprises:
  • the cells are combined into blocks, the gradient direction histogram is normalized within the block, and the gradient direction histograms of all blocks in the fingerprint image are combined to form an HOG feature.
  • the determining, according to the HOG feature and the optimal classification surface, the quality of the fingerprint image comprises:
  • the invention also provides a device for judging the quality of a fingerprint image, comprising a learning module and a judging module, wherein:
  • a learning module configured to acquire a fingerprint image sample, and use an SVM classifier to learn according to the fingerprint image sample to obtain an optimal classification surface
  • the determining module is configured to obtain a fingerprint image to be determined, calculate a HOG gradient direction histogram feature of the fingerprint image, and determine a quality of the fingerprint image according to the HOG feature and an optimal classification surface.
  • the learning module is configured to: calculate an HOG feature of the fingerprint image sample, input the HOG feature of the fingerprint image sample into the SVM classifier for training learning, and obtain an optimal classification surface.
  • the determining module includes a processing unit, and the processing unit is configured to: calculate the a gradient direction value of each pixel position in the fingerprint image; dividing the fingerprint image into a plurality of cells, constructing a gradient direction histogram according to the gradient direction value for each cell; combining the cells into blocks, The gradient direction histogram is normalized in the block, and the gradient direction histograms of all the blocks in the fingerprint image are combined to form an HOG feature.
  • the determining module comprises a processing unit and a discriminating unit, wherein:
  • the determining unit is configured to determine according to the calculation result of the processing unit, and if the calculation result is f(x)>0, it is determined that the quality of the fingerprint image is good; if the calculation result is f(x) ⁇ 0, the determining is performed. The quality of the fingerprint image is poor.
  • the invention provides a method for judging the quality of a fingerprint image, and the SVM classifier is used to learn the fingerprint image sample to obtain an optimal classification surface, and introduces the HOG feature into the fingerprint image quality judgment according to the HOG feature and the optimal classification surface. Automatically judge the quality of the fingerprint image. It not only saves the work of manually determining the judgment threshold, but also has a good expansion ability, that is, this judgment method can determine the influence of various types of noise, and only need to input the required sample type to complete the judgment, in a large number of experiments It shows excellent results.
  • the judging method of the invention is mainly applicable to the judgment of the quality of the collected fingerprint image by the small-sized fingerprint collecting device, especially for the situation that the fingerprint image is blurred caused by sweat, muddy, noise, etc., and the quality of the fingerprint image before the fingerprint recognition Making accurate judgments lays a good foundation for improving fingerprint recognition rate and reducing falsehood rate.
  • FIG. 1 is a flow chart of an embodiment of a method for judging fingerprint image quality according to the present invention
  • FIG. 2 is a flow chart of calculating a HOG feature of a fingerprint image sample in an embodiment of the present invention
  • FIG. 3 is a schematic diagram of an optimal classification plane in an embodiment of the present invention.
  • FIG. 4 is a flowchart of calculating a HOG feature of a fingerprint image in an embodiment of the present invention
  • FIG. 5 is a structural block diagram of an embodiment of a fingerprint image quality determining apparatus according to the present invention.
  • FIG. 6 is a structural block diagram of a judging module in an embodiment of the present invention.
  • the method for judging the quality of the fingerprint image of the present invention takes into account the HOG (Histogram of Oriented Gradient) feature to describe the image gradient direction distribution, which is an effective texture statistical feature. Therefore, by introducing the HOG feature and using the SVM to supervise the learning mode. To successfully distinguish the quality of the fingerprint image, the accurate determination of the fingerprint image quality in the small-size fingerprint acquisition device is realized.
  • HOG Heistogram of Oriented Gradient
  • the determining method includes the following steps:
  • Step S101 Acquire a fingerprint image sample
  • the fingerprint image sample includes positive and negative samples, that is, a good quality sample and a poor quality sample, and at least one positive and negative sample, preferably a plurality.
  • the fingerprint image sample is manually selected, and may be a fingerprint image acquired manually on site, or may be an off-the-shelf fingerprint image obtained from the outside.
  • Step S102 learning by using the SVM classifier according to the fingerprint image sample to obtain an optimal classification surface
  • the HOG feature of the fingerprint image sample is first calculated, and then the HOG feature of the fingerprint image sample is input into a SVM (Support Vector Machine) classifier for training learning, thereby obtaining an optimal classification surface.
  • SVM Small Vector Machine
  • the HOG feature is a feature descriptor for object detection, which constructs features by statistically analyzing the gradient direction histogram of the local region of the image.
  • the gradient direction histogram describes the image gradient direction distribution and is an effective texture statistical property.
  • the calculation method of the HOG feature of the fingerprint image sample is as shown in FIG. 2, and includes the following steps:
  • Step S121 Calculating a gradient direction value of each pixel position in the fingerprint image sample
  • the gradient of the horizontal and vertical coordinate directions in the fingerprint image sample is calculated, and the gradient direction value of each pixel position is calculated.
  • the gradient at the pixel point (x, y) in the fingerprint image sample is:
  • G x (x, y), G y (x, y), and I(x, y) represent horizontal gradients, vertical gradients, and image gray values at pixel points (x, y) in the fingerprint image sample, respectively.
  • the gradient magnitude G(x, y) and the gradient direction ⁇ (x, y) at the pixel (x, y) are:
  • Step S122 dividing the fingerprint image sample into a plurality of cells, and constructing a gradient direction histogram for each cell according to the gradient direction value.
  • a histogram of 9 bins is used to count the gradient information of the 6*6 pixels, that is, the gradient direction of the cell is divided into three degrees by 360 degrees.
  • Direction block For example, if the gradient direction value of this pixel is 20-40 degrees and the gradient magnitude is 2, the second bin of the histogram has a count of 2, so that each pixel in the cell is weighted by a gradient direction histogram ( Map to a fixed angular range), you can get the gradient direction histogram of this cell, which is the 9-dimensional feature vector corresponding to the cell (because there are 9 bins).
  • Step S123 Combine the cells into blocks, and normalize the gradient direction histogram in the block.
  • each cell is combined into a larger, spatially connected interval to form a block, such that the feature vectors of all cells in a block are concatenated to obtain the HOG feature of the block.
  • These intervals overlap each other, which means that the characteristics of each cell appear multiple times in the last feature vector with different results.
  • the block descriptor we will normalize is called the HOG descriptor.
  • Step S124 Combine the gradient direction histograms of all the blocks in the fingerprint image sample to form the HOG feature.
  • a detection window is defined locally in the fingerprint image sample, and only the detection of the HOG feature is performed on the detection window.
  • the entire fingerprint image sample is scanned step by step in a manner of detecting a window to obtain an HOG feature of the entire fingerprint image sample.
  • the SVM classifier After calculating the HOG feature of the fingerprint image sample, the SVM classifier then performs training learning according to the HOG features of the plurality of fingerprint image samples to obtain an optimal classification surface.
  • SVM is a very classic supervised learning method in the field of machine learning (that is, manual determination of samples). It has been widely used in text classification and other fields, and has achieved good results.
  • Fingerprint image quality analysis can be regarded as a two-category problem, that is, the quality is +1 and the quality difference is -1.
  • the SVM classifier needs to find a linear hyperplane in the high-dimensional space, which will be subordinate Separate data points in two categories.
  • the two-dimensional plane there are two types of points, white and black. It is necessary to find a plane to separate the two types of points. According to the situation in the figure, the plane can find an infinite number, and the SVM classifier takes the plane farthest from the boundary of the two types of data points as the optimal classification plane, that is, the minimum distance between the two types of data points to the plane is the largest.
  • a support hyperplane is defined, that is, planes H1 and H2 indicated by two broken lines in the figure.
  • the parameters w and b of the optimal classification plane are obtained.
  • the parameter w of the optimal classification plane is obtained by linear summation of the sample points, and these sample points must fall between two supporting hyperplanes, which are called support vectors.
  • the weights of the sample points outside the two supporting hyperplanes are all 0, which has no effect on determining the optimal hyperplane.
  • the optimal classification plane obtained by the SVM classifier is only affected by a part of the data, and the performance is relatively stable, and the hyperplane obtained by maximizing the interval tends to have better generalization performance, that is, classifying the unknown data. Can have a lower error rate.
  • the quality of the fingerprint image can be judged.
  • Step S103 Acquire a fingerprint image to be determined
  • the fingerprint image may be a fingerprint image acquired in the field, or may be a fingerprint image acquired from the outside.
  • Step S104 Calculating the HOG feature of the fingerprint image
  • the specific calculation process of the HOG feature of the fingerprint image is as shown in FIG. 4, and includes the following steps:
  • Step S141 Calculating a gradient direction value of each pixel position in the fingerprint image
  • Step S142 Dividing the fingerprint image into a plurality of cells, and constructing a gradient direction histogram for each cell according to the gradient direction value.
  • Step S143 Combine the cells into blocks, and normalize the gradient direction histogram in the block.
  • Step S144 Combine the gradient direction histograms of all the blocks in the fingerprint image to form the HOG feature.
  • the method for calculating the HOG feature of the fingerprint image in the step S104 is the same as the method for calculating the HOG feature of the fingerprint image sample in step S102, and details are not described herein again.
  • Step S105 determining the quality of the fingerprint image according to the HOG feature and the optimal classification surface
  • the parameter constant term b is 2.8326173060507497e+001
  • the support vector w is as follows, which is a 1296-dimensional data:
  • HOG feature x of a fingerprint image is calculated as follows, which is a 1296-dimensional data:
  • HOG feature x of another fingerprint image is calculated as follows, which is a 1296-dimensional data:
  • the method for judging the quality of the fingerprint image of the present invention determines the optimal classification surface by means of machine learning by introducing the HOG feature into the fingerprint image quality judgment, and determines the quality of the fingerprint image according to the HOG feature of the fingerprint image and the optimal classification surface. Good or bad. It not only saves the work of manually determining the judgment threshold, but also has a good expansion ability. That is, this judgment method can determine the influence of various types of noise, and only need to input the required sample type to complete the judgment, in a large number of experiments. Shows excellent results.
  • the judging method of the invention is mainly applied to the small-sized fingerprint collecting device to judge the quality of the collected fingerprint image, especially for the case that the fingerprint image is blurred due to sweat, muddy, noise, etc., and the quality of the fingerprint image before the fingerprint recognition Making accurate judgments lays a good foundation for improving fingerprint recognition rate and reducing falsehood rate.
  • the determining apparatus includes a learning module and a determining module.
  • Learning module used to acquire fingerprint image samples, use SVM classifier to learn according to fingerprint image samples, obtain the optimal classification surface, and send the function of the optimal classification surface and related parameters to the judgment module.
  • the fingerprint image sample includes positive and negative samples, that is, a good quality sample and a poor quality sample, and at least one positive and negative sample, preferably a plurality.
  • the fingerprint image sample is manually selected, and may be a fingerprint image acquired manually on site, or may be an off-the-shelf fingerprint image obtained from the outside.
  • the learning module first calculates the HOG feature of the fingerprint image sample, and then inputs the HOG feature of the fingerprint image sample into the SVM classifier for training learning, thereby obtaining an optimal classification surface.
  • the learning module When calculating the HOG feature, the learning module first calculates the gradient direction value of each pixel position in the fingerprint image sample; then divides the fingerprint image sample into a plurality of cells, and constructs a gradient direction histogram according to the gradient direction value for each cell; The cells are combined into blocks, and the gradient direction histogram is normalized in the block; finally, the gradient direction histograms of all the blocks in the fingerprint image sample are combined to form an HOG feature.
  • the learning module inputs the HOG features of the plurality of fingerprint image samples into the SVM, and the SVM performs training learning according to the HOG features of the plurality of fingerprint image samples to obtain an optimal classification surface.
  • the judging module is configured to obtain a fingerprint image to be judged, calculate a HOG feature of the fingerprint image, determine a quality of the fingerprint image according to the HOG feature and the optimal classification surface, and output a judgment result.
  • the judgment module includes a processing unit and a discriminating unit, wherein:
  • the processing unit is configured to obtain a fingerprint image to be determined, calculate an HOG feature of the fingerprint image, input the HOG feature into a function of an optimal classification surface, and send the determination result to the determining unit.
  • the processing unit first calculates a gradient direction value of each pixel position in the fingerprint image; then divides the fingerprint image into a plurality of cells, and constructs a gradient direction histogram according to the gradient direction value; and then combines the cells into Block, normalize the gradient direction histogram within the block, and combine the gradient direction histograms of all the blocks in the fingerprint image to form the HOG feature.
  • the fingerprint image quality judging device of the present invention adopts the HOG feature and adopts the SVM.
  • the supervised learning mode successfully separates the fingerprint image quality, especially for the case where the fingerprint image is blurred due to sweat, mud, noise, etc., and can accurately judge the quality of the fingerprint image before the fingerprint recognition, in order to improve the recognition rate and reduce the falsehood.
  • the rate lays a good foundation, especially for small-size fingerprint acquisition devices.
  • the judging device of the invention not only saves the work of manually determining the judgment threshold, but also has a good expansion capability, that is, the judging method can determine the influence of various types of noise, and only needs to input the required sample type to complete Judging, it showed excellent results in a large number of experiments.
  • the device for determining the quality of the fingerprint image provided by the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the function may be allocated according to needs. Different functional modules are completed.
  • the device for determining the quality of the fingerprint image provided by the above embodiment is the same as the embodiment of the method for determining the quality of the fingerprint image.
  • the specific implementation process is described in the method embodiment, and the technical features in the method embodiment are in the device embodiment. Correspondence is applicable, and will not be described here.
  • the invention provides a method for judging the quality of a fingerprint image, and the SVM classifier is used to learn the fingerprint image sample to obtain an optimal classification surface, and introduces the HOG feature into the fingerprint image quality judgment according to the HOG feature and the optimal classification surface. Automatically judge the quality of the fingerprint image. It not only saves the work of manually determining the judgment threshold, but also has a good expansion ability, that is, this judgment method can determine the influence of various types of noise, and only need to input the required sample type to complete the judgment, in a large number of experiments It shows excellent results.
  • the judging method of the invention is mainly applicable to the judgment of the quality of the collected fingerprint image by the small-sized fingerprint collecting device, especially for the situation that the fingerprint image is blurred caused by sweat, muddy, noise, etc., and the quality of the fingerprint image before the fingerprint recognition Making accurate judgments lays a good foundation for improving fingerprint recognition rate and reducing falsehood rate.

Abstract

本发明公开了一种指纹图像质量的判断方法和装置，所述判断方法包括步骤：获取指纹图像样本；利用SVM分类器根据所述指纹图像样本进行学习，获得最优分类面；获取待判断的指纹图像，并计算所述指纹图像的HOG特征；根据所述HOG特征和最优分类面判断所述指纹图像的质量。从而实现了在小尺寸指纹采集设备中对指纹图像质量的准确判断，不仅省去人工确定判断阈值的工作，而且具有很好的扩展能力，即这种判断方式可以判定多种类型噪声带来的影响，只需输入需要的样本类型即可完成判断，在大量实验中表现出极佳的效果，为提高指纹识别率、降低认假率打下了良好的基础。

Description

指纹图像质量的判断方法和装置 技术领域
本发明涉及通信技术领域，尤其是涉及一种指纹图像质量的判断方法和装置。
背景技术
指纹识别技术已开始广泛应用于移动终端，针对移动终端的指纹识别，识别算法则是核心技术。由于指纹采集传感器的限制，当手指带有汗渍、泥渍等情况下，识别率会大大降低，特别是移动终端这种小尺寸指纹采集设备，识别算法更是依赖输入的指纹图像质量，质量越高则识别率越高，认假率越低，因此在识别之前对指纹图像质量的判断尤为重要。
传统的判断方法，首先统计指纹图像的均值、方差、信息熵、环形谱结构等特征指标，然后综合这些指标计算质量分值。但这类方法只适用于全采集指纹识别系统，而针对小尺寸指纹采集系统并不适用；再者这些特征指标也并没有充分考虑指纹是具有特定纹理结构的图像。因此，现有技术中的指纹图像质量的判断方法，还不能对小尺寸指纹采集设备(或系统)中指纹图像质量进行准确的判断。
发明内容
本发明的主要目的在于提供一种指纹图像质量的判断方法和装置，旨在自动判断指纹图像质量，为提高指纹识别率、降低认假率打下良好的基础。
为达以上目的，本发明提出一种指纹图像质量的判断方法，包括步骤：
获取指纹图像样本；
利用SVM支持向量机分类器根据所述指纹图像样本进行学习，获得最优分类面；
获取待判断的指纹图像，并计算所述指纹图像的HOG梯度方向直方图特征；
根据所述HOG特征和最优分类面判断所述指纹图像的质量。
优选地，所述利用SVM支持向量机分类器根据所述指纹图像样本进行学习获得最优分类面包括：
计算所述指纹图像样本的HOG特征；
将所述指纹图像样本的HOG特征输入所述SVM分类器中进行训练学习，获得最优分类面。
优选地，所述计算所述指纹图像的HOG梯度方向直方图特征包括：
计算所述指纹图像中每一像素位置的梯度方向值；
将所述指纹图像分成多个单元格，根据所述梯度方向值为每个单元格构建梯度方向直方图；
将所述单元格组合成块，在所述块内归一化所述梯度方向直方图，将所述指纹图像中所有块的梯度方向直方图结合起来形成HOG特征。
优选地，所述根据所述HOG特征和最优分类面判断所述指纹图像的质量包括：
将所述指纹图像的HOG特征输入所述最优分类面的函数f(x)＝w*x-b中进行计算，其中w为最优分类面的支持向量，b为常数项，x为指纹图像的HOG特征；
若计算结果为f(x)＞0，则判定所述指纹图像的质量好；
若计算结果为f(x)＜0，则判定所述指纹图像的质量差。
本发明同时提出一种指纹图像质量的判断装置，包括学习模块和判断模块，其中：
学习模块，设置为获取指纹图像样本，利用SVM分类器根据所述指纹图像样本进行学习，获得最优分类面；
判断模块，设置为获取待判断的指纹图像，并计算所述指纹图像的HOG梯度方向直方图特征，根据所述HOG特征和最优分类面判断所述指纹图像的质量。
优选地，所述学习模块设置为：计算所述指纹图像样本的HOG特征，将所述指纹图像样本的HOG特征输入所述SVM分类器中进行训练学习，获得最优分类面。
优选地，所述判断模块包括处理单元，所述处理单元设置为：计算所述 指纹图像中每一像素位置的梯度方向值；将所述指纹图像分成多个单元格，根据所述梯度方向值为每个单元格构建梯度方向直方图；将所述单元格组合成块，在所述块内归一化所述梯度方向直方图，将所述指纹图像中所有块的梯度方向直方图结合起来形成HOG特征。
优选地，所述判断模块包括处理单元和判别单元，其中：
处理单元，设置为将所述指纹图像的HOG特征输入所述最优分类面的函数f(x)＝w*x-b中进行计算，其中w为最优分类面的支持向量，b为常数项，x为指纹图像的HOG特征；
判别单元，设置为根据处理单元的计算结果进行判别，若计算结果为f(x)＞0，则判定所述指纹图像的质量好；若计算结果为f(x)＜0，则判定所述指纹图像的质量差。
本发明所提供的一种指纹图像质量的判断方法，通过SVM分类器对指纹图像样本进行学习获得最优分类面，并将HOG特征引入到指纹图像质量判断中，根据HOG特征和最优分类面自动判断指纹图像的质量。不仅省去了人工确定判断阈值的工作，而且具有很好的扩展能力，即这种判断方式可以判定多种类型噪声带来的影响，只需输入需要的样本类型即可完成判断，在大量实验中表现出极佳的效果。本发明的判断方法主要适用于小尺寸指纹采集设备对采集的指纹图像质量好坏的判断，尤其针对汗渍、泥渍、噪声等造成指纹图像模糊的情形，可以在指纹识别前对指纹图像的质量进行准确的判断，为提高指纹识别率、降低认假率打下了良好的基础。
附图说明
图1是本发明指纹图像质量的判断方法一实施例的流程图；
图2是本发明实施例中计算指纹图像样本的HOG特征的流程图；
图3是本发明实施例中最优分类面的示意图；
图4是本发明实施例中计算指纹图像的HOG特征的流程图；
图5是本发明指纹图像质量的判断装置一实施例的结构框图；
图6是本发明实施例中判断模块的结构框图。
本发明目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
本发明的指纹图像质量的判断方法，考虑到HOG(Histogram of Oriented Gradient，梯度方向直方图)特征描述了图像梯度方向分布，是有效的纹理统计特征，因此通过引入HOG特征并采用SVM监督学习模式来成功的区分出指纹图像质量，实现了对小尺寸指纹采集设备中指纹图像质量进行准确的判断。
参见图1，提出本发明的指纹图像质量的判断方法一实施例，所述判断方法包括以下步骤：
步骤S101：获取指纹图像样本
指纹图像样本包括正负样本，即质量好的样本和质量差的样本，正负样本至少各一个，优选多个。指纹图像样本由人工选择，可以是现场人工采集获取的指纹图像，也可以是从外部获取的现成的指纹图像。
步骤S102：利用SVM分类器根据指纹图像样本进行学习，获得最优分类面
具体的，首先计算指纹图像样本的HOG特征，然后将指纹图像样本的HOG特征输入SVM(Support Vector Machine，支持向量机)分类器中进行训练学习，从而获得最优分类面。
HOG特征是一种物体检测的特征描述子，它通过统计图像局部区域的梯度方向直方图来构成特征。梯度方向直方图描述了图像梯度方向分布，是有效的纹理统计特性。指纹图像样本的HOG特征的计算方法如图2所示，包括以下步骤：
步骤S121：计算指纹图像样本中每一像素位置的梯度方向值
具体的，计算指纹图像样本中横纵坐标方向的梯度，并根此计算每个像素位置的梯度方向值。指纹图像样本中像素点(x,y)处的梯度为:
Gx(x，y)＝I(x+1，y)-I(x-1，y)
Gy(x，y)＝I(X，y+1)-I(X，y-1)
其中Gx(x,y)、Gy(x,y)和I(x,y)分别表示指纹图像样本中像素点(x,y)处的水平方向梯度、垂直方向梯度和图像灰度值。像素点(x,y)处的梯度幅值G(x,y)和梯度方向θ(x,y)分别为：
Figure PCTCN2014094567-appb-000001
Figure PCTCN2014094567-appb-000002
步骤S122：将指纹图像样本分成多个单元格，根据梯度方向值为每个单元格构建梯度方向直方图
具体的，假设每个单元格(cell)为6*6个像素，采用9个bin的直方图来统计这6*6个像素的梯度信息，也就是将单元格的梯度方向360度分成9个方向块。例如：如果这个像素的梯度方向值是20-40度，梯度幅值是2，则直方图第二个bin的计数就是2，这样对单元格内每个像素用梯度方向直方图进行加权投影(映射到固定的角度范围)，就可以得到这个单元格的梯度方向直方图，也即该单元格对应的9维特征向量(因为有9个bin)。
步骤S123：将单元格组合成块，在块内归一化梯度方向直方图
具体的，将各个单元格组合成更大的、空间上连通的区间而形成块(block)，这样，一个块内所有单元格的特征向量串联起来便得到该块的HOG特征。这些区间互相重叠，这就意味着：每一个单元格的特征会以不同的结果多次出现在最后的特征向量中。我们将归一化之后的块描述符就称之为HOG描述符。
步骤S124：将指纹图像样本中所有块的梯度方向直方图结合起来形成HOG特征
最后将指纹图像样本中所有重叠块进行HOG特征收集，并将它们结合成最终的特征向量，即指纹图像样本的HOG特征。
在某些实施例中，在指纹图像样本的局部限定一个检测窗口，只对该检测窗口进行HOG特征的计算。或者，以检测窗口的方式逐步对整个指纹图像样本进行扫描获得整个指纹图像样本的HOG特征。
计算出指纹图像样本的HOG特征后，接下来，SVM分类器则根据多个指纹图像样本的HOG特征进行训练学习，获得最优分类面。SVM是机器学习领域非常经典的有监督学习方法(即人工确定样本)，目前已广泛应用于文本分类等领域中，取得了很好的效果。
指纹图象质量分析可以看作二分类问题，即质量好为+1，质量差为-1。对于二分类问题，SVM分类器需要在高维空间中找到一个线性超平面，将分属 于两个类别的数据点分开。以图3为例，在二维平面中，有白色和黑色两类点，需要找到一个平面将两类点分开。按图中的情形，平面可以找到无限个，而SVM分类器把离两类数据点边界最远的平面作为最优分类面，即使得两类数据点到平面的最小距离最大。
具体实现上，可以定义一个超平面作为最优分类面,即图中的实线表示的平面，其使得标签值y＝-1的点落在f(x)<0的一侧，而y＝+1的点落在f(x)>0的一侧。定义最优分类面的函数为f(x)＝w·x-b，其中w为最优分类面的支持向量，b为常数项，x为输入的图像的HOG特征，w和b为最优分类面的参数。此外再定义一个支持超平面，即图中两条虚线表示的平面H1和H2。假定|(w,b)|＝1，则支持超平面的公式为wx-b＝±1。根据SVM的假设，对于最优分类面，所有数据点应该满足y(wx-b)≥1。同时可以推出，两个支持超平面之间的距离d＝2/|w|。由此可以得到SVM的目标函数：
Figure PCTCN2014094567-appb-000003
对上述问题利用拉格朗日乘子法求解，可以得到：
Figure PCTCN2014094567-appb-000004
最终获得最优分类面的参数w和b。由上面各个式子可以看出，最优分类面的参数w是由样本点线性加和得到的，而且这些样本点必然落在两个支持超平面之间，这些样本点被称为支持向量。而在两个支持超平面之外的样本点的权重均为0，对决定最优超平面没有任何影响。
因此，SVM分类器求得的最优分类面只受一部分数据的影响，性能就相对稳定，而且通过最大化间隔求得的超平面往往具有较好的泛化性能，即在给未知数据做分类的时候能有较低的错误率。
当通过自动训练学习获得最优分类面后，则可对指纹图像的质量进行判断。
步骤S103：获取待判断的指纹图像
该指纹图像可以是现场采集的指纹图像，也可以是从外部获取的指纹图像。
步骤S104：计算指纹图像的HOG特征
指纹图像的HOG特征的具体计算流程如图4所示，包括以下步骤：
步骤S141：计算指纹图像中每一像素位置的梯度方向值
步骤S142：将指纹图像分成多个单元格，根据梯度方向值为每个单元格构建梯度方向直方图
步骤S143：将单元格组合成块，在块内归一化梯度方向直方图
步骤S144：将指纹图像中所有块的梯度方向直方图结合起来形成HOG特征
本步骤S104中指纹图像的HOG特征的计算方法，与步骤S102中指纹图像样本的HOG特征的计算方法相同，在此不再赘述。
步骤S105：根据HOG特征和最优分类面判断指纹图像的质量
具体的，将指纹图像的HOG特征输入最优分类面的函数f(x)＝w*x-b中进行计算，其中w为最优分类面的支持向量，b为常数项，x为指纹图像的HOG特征。如果计算结果为f(x)＞0，则判定指纹图像的质量好，输出标签值y＝+1；如果计算结果为f(x)＜0，则判定指纹图像的质量差，输出标签值y＝-1。
举例而言，假设在学习样本过后获得最优分类面，其参数常数项b为2.8326173060507497e+001，支持向量w如下所示，为一1296维数据：
-5.65062046e-001 -1.22261524e-001 6.23945475e-001 2.61808070e-003 -5.96388519e-001 7.09277689e-001 -1.10982299e+000 2.81930566e-001 1.25543416e+000 -2.58267093e+000 -3.93080682e-001 1.26033258e+000 7.85001144e-002 -3.78241152e-001 7.57321477e-001 2.73346394e-001 2.01390719e+000 4.89104331e-001 6.39280975e-001 5.81095874e-001 2.78728080e+000 -2.40852520e-001 -1.01792566e-001 1.00278020e+000 1.44761455e+000 -1.58462083e+000 1.89574575e+000 1.07644367e+000 8.66438568e-001 -6.86612964e-001 9.07730460e-001 -3.49271774e-001 1.22522628e-002 -6.30278111e-001 3.00675720e-001 1.85192859e+000 1.62126279e+000 4.82157320e-001 -1.34164858e+000 2.04924569e-001  1.55832696e+000 1.77290034e+000 -5.20038784e-001 7.18480110e-001 2.09454685e-001 7.86378205e-001 -1.78212130e+000 9.68096316e-001 8.70740235e-001 -6.44824266e-001 2.54506993e+000 7.10121095e-001 -5.98418236e-001 -1.06928796e-001 8.71971488e-001 1.17480731e+000 7.15802729e-001 2.14615989e+000 -2.88301253e+000 -1.92418182e+000 2.96458960e-001 -8.55592549e-001 -2.03400755e+000 -6.68073654e-001 1.38882890e-001 -7.04979658e-001 -1.46523386e-001 1.70489192e+000 -1.59027326e+000 -1.56655461e-001 -1.12606674e-001 3.60005188e+000 -2.28805041e+000 1.52749741e+000 1.74320495e+000 -6.91233456e-001 2.85730386e+000 -1.05900079e-001 -1.68506277e+000 -1.18375808e-001 -1.10159695e+000 6.55481517e-001 -8.89651656e-001 5.75935364e-001 -8.52293670e-001 -1.37630284e+000 7.58536041e-001 1.52515018e+000 6.06581271e-001 1.50268352e+000 1.77640891e+000 -7.86289200e-002 1.14980662e+000 1.16938233e+000 2.01216507e+000 -6.11184180e-001 -7.25955606e-001 -2.55225211e-001 7.07862496e-001 4.88763779e-001 7.70018041e-001 -1.96791291e+000 2.75504446e+000 -9.32191432e-001 -2.94734448e-001 6.88520610e-001 -5.86638331e-001 1.37036932e+000 -1.00271665e-001 1.41746268e-001 -2.22995734e+000 2.95861661e-001 2.21844387e+000 1.66081011e+000 -4.16236019e+000 4.55661625e-001 1.07015693e+000 2.21201491e+000 -1.26312482e+000 -9.47408438e-001 2.54873562e+000 1.56559813e+000 -1.24778640e+000 -1.32051051e+000 1.65815008e+000 -1.41459537e+000 1.92478895e+000 -1.84008098e+000 3.61277342e+000 2.20739937e+000 -1.34330928e+000 -5.74071646e-001 1.62802529e+000 7.33614326e-001 -2.55355924e-001 -1.70272529e+000 1.97496819e+000 -1.09231484e+000 -2.04441524e+000 -3.81036431e-001 2.97871494e+000 -3.68865132e+000 3.15781379e+000 5.97642064e-001 5.80903649e-001 -2.13103652e+000 -8.67208302e-001 1.80704606e+000 9.69478860e-002 4.28946495e-001 5.80691278e-001 -5.95060766e-001 9.76634204e-001 -7.90543929e-002 -4.92165685e-001 1.70432973e+000 -1.05518484e+000 8.51133049e-001 -1.17313042e-001 1.35035202e-001 2.53540706e-002 6.81154847e-001 -1.51302606e-001 -1.07936570e-002 -1.31098163e+000 2.20081162e+000 -3.33757579e-001 1.28789783e+000 -1.38676023e+000 -5.06396294e-001 1.09426820e+000 3.83968383e-001 1.24972117e+000 2.32150960e+000 5.33086717e-001 1.50917560e-001 5.07387817e-001 -1.61430746e-001 2.17941809e+000 3.69923621e-001  2.83043933e+000 1.90740168e-001 1.46296895e+000 1.37301970e+000 6.74435854e-001 -6.71533048e-001 -9.91295278e-001 9.93396699e-001 -3.25666308e-001 1.57782927e-001 -1.24856126e+000 -1.44502115e+000 8.13267171e-001 1.03857972e-001 -2.26438427e+000 1.86400020e+000 1.52461588e+000 2.44749278e-001 1.56721997e+000 -1.32312346e+000 -8.10653090e-001 -2.48049855e+000 2.99090832e-001 5.61489940e-001 -2.98761845e-001 -3.92122388e+000 -1.46208012e+000 -2.30779409e+000 -5.69158137e-001 9.63538647e-001 1.94473159e+000 -8.96440353e-003 1.95122707e+000 -2.46136642e+000 2.12227440e+000 2.57141161e+000 -1.22917616e+000 2.47263566e-001 -1.08239722e+000 4.82539266e-001 -1.97637364e-001 -4.98678893e-001 -7.77484000e-001 1.73539460e+000 4.62011158e-001 6.98073208e-003 7.24058747e-001 -3.12087923e-001 1.10251451e+000 1.25460291e+000 1.75949061e+000 -5.29235363e-001 -9.64799151e-002 -2.53143954e+000 -4.57035750e-001 3.02866429e-001 1.37184167e+000 3.23770791e-001 -2.42519689e+000 5.12698852e-002 -8.26845348e-001 -1.34545541e+000 -8.93717766e-001 1.43910718e+000 9.63344097e-001 -8.71918917e-001 -1.07027996e+000 1.02522659e+000 1.23496354e+000 4.57389086e-001 1.19853270e+000 1.64693880e+000 5.11548460e-001 -1.33672154e+000 2.71858722e-001 -1.17530978e+000 6.15254566e-002 2.94417411e-001 4.36541080e-001 1.37163055e+000 -1.82581842e+000 -6.85078382e-001 -7.76390493e-001 -1.97848654e+000 2.16851807e+000 1.60647643e+000 -1.86727822e-001 6.96594656e-001 -7.91485071e-001 2.65202731e-001 1.50661623e+000 1.07845032e+000 1.78456712e+000 -1.79334183e-003 1.40141740e-001 1.53397667e+000 1.23211336e+000 7.10765898e-001 5.86186089e-002 2.19317412e+000 2.13756412e-001 3.66310835e-001 5.48000991e-001 -6.51535928e-001 -1.14362442e+000 -4.40639943e-001 1.27369392e+000 1.44236636e+000 1.37426305e+000 2.02852416e+000 -3.16964775e-001 -2.62251306e+000 4.59316635e+000 -1.96874535e+000 8.52879345e-001 -2.99564815e+000 7.21852005e-001 9.10931945e-001 -2.10801840e+000 -8.83882880e-001 1.07756746e+000 -1.12351263e+000 3.39177704e+000 1.30952287e+000 3.86621094e+000 3.64660531e-001 1.86861193e+000 1.12214160e+000 1.42694950e-001 -6.56762421e-001 -6.42201722e-001 7.56536007e-001 -7.53396690e-001 1.51442754e+000 -1.55608511e+000 -3.20019364e-001 1.42415535e+000 -1.35575104e+000 7.34291017e-001 1.73662949e+000  -9.04838800e-001 -1.99635935e+000 -1.95133519e+000 -8.14499021e-001 3.27238366e-002 3.87542397e-001 -1.95944226e+000 -1.91607440e+000 2.20693421e+000 1.58080161e+000 -3.00249863e+000 -1.07961583e+000 -6.12612069e-001 -9.43629518e-002 2.07135153e+000 -1.07898617e+000 1.98396698e-001 -8.05510044e-001 9.38360572e-001 -1.17815860e-001 -1.42652404e+000 -1.76703289e-001 -4.94757183e-002 -1.07781088e+000 8.66706133e-001 -1.95579135e+000 -1.15027654e+000 1.17117381e+000 -7.06813395e-001 1.68010008e+000 6.21922314e-001 1.40011585e+000 -5.73635876e-001 5.50615430e-001 3.30518556e+000 3.77227396e-001 7.73974836e-001 3.53976369e+000 1.18661797e+000 1.13880858e-001 1.61683464e+000 3.84422779e-001 -7.32956529e-001 -2.02169132e+000 2.86095762e+000 -5.44128776e-001 1.94903874e+000 3.54368895e-001 9.16079760e-001 1.09807134e+000 -8.51657987e-002 -8.94440413e-002 1.56543106e-001 -7.25803673e-001 -2.16020489e+000 9.41323280e-001 -2.19787073e+000 1.78099692e+000 1.21620715e+000 1.99195230e+000 -1.74088085e+000 -1.65861666e+000 1.76478863e+000 2.54104710e+000 -1.49221981e+000 5.00257432e-001 8.48669946e-001 -1.79053199e+000 -2.45864248e+000 2.42074177e-001 1.22374451e+000 -8.02697659e-001 -1.12856038e-001 -1.45404863e+000 -1.14215183e+000 -7.71217883e-001 5.67980647e-001 1.74748570e-001 1.29376799e-001 -9.47380960e-001 -1.38257396e+000 1.08610772e-001 -2.31834769e+000 -3.18607867e-001 5.97291589e-001 2.97453463e-001 -2.88981467e-001 2.55147290e+000 4.35854554e-001 1.92910707e+000 -4.04442739e+000 3.35437202e+000 -7.18500257e-001 1.71847641e+000 3.76633972e-001 7.17979133e-001 -2.29244590e+000 4.33288908e+000 -3.21098715e-001 4.49396563e+000 -1.51703286e+000 -8.13243508e-001 3.68480325e-001 -2.04157472e+000 2.64199209e+000 -9.59263504e-001 -7.27866054e-001 -9.23123717e-001 1.79922330e+000 5.94993055e-001 1.60790789e+000 -1.55783963e+000 1.17599356e+000 3.75915971e-003 5.80935240e-001 -6.17859364e-001 5.47859287e+000 1.81236637e+000 -6.58016875e-002 1.29181659e+000 2.71043873e+000 1.00019448e-001 1.61559093e+000 1.83883631e+000 2.07489991e+000 8.18653643e-001 -7.93111801e-001 1.19132567e-002 4.03664291e-001 -1.44269729e+000 1.02415705e+000 2.56434369e+000 -7.33990967e-001 2.51126194e+000 -2.70398140e+000 -1.83867002e+000 1.18612683e+000 1.02612591e+000 -1.12655747e+000 -1.31735340e-001  -1.88770008e+000 -1.63585925e+000 -1.10935783e+000 1.27429748e+000 -5.52048720e-002 3.98856848e-001 -1.72617018e-001 -1.92361295e+000 -4.13408369e-001 -1.13245070e+000 2.85943294e+000 -2.07407784e+000 2.00722980e+000 7.59493232e-001 7.53989220e-001 5.31222701e-001 -1.06769693e+000 -7.28021516e-003 4.78438735e-001 -1.25011992e+000 -2.75211215e-001 -1.55662167e+000 -9.34187949e-001 -4.09402192e-001 1.53119981e+000 -1.25467312e+000 1.53636479e+000 -7.30930805e-001 2.75530076e+000 2.67432421e-001 2.85938501e+000 -1.42511964e+000 7.33835816e-001 1.99697030e+000 -3.78929794e-001 3.14493728e+000 1.20242715e+000 -2.51046824e+000 8.01128805e-001 -7.64064312e-001 1.19065619e+000 -7.97462463e-001 -4.37173009e-001 -3.08697790e-001 -2.21754000e-001 -2.32537538e-001 1.42117727e+000 1.26279199e+000 8.76116455e-001 -5.89745462e-001 -7.15154827e-001 -1.47839296e+000 3.58165920e-001 -1.45312059e+000 -1.79811522e-001 -1.22732484e+000 1.35544467e+000 -1.39437079e+000 4.28034633e-001 -4.00926471e-001 -4.24189746e-001 -1.87310791e+000 -7.61907935e-001 1.94510496e+000 -6.93394570e-003 -1.69082570e+000 1.55784622e-001 1.70695126e+000 1.48444736e+000 1.78316402e+000 6.74396753e-001 2.54230237e+000 3.12799782e-001 -2.27586889e+000 1.64418113e+000 -5.52159488e-001 1.09172642e+000 -1.33025026e+000 4.27963167e-001 -1.06191821e-002 2.02915812e+000 1.52300692e+000 1.75060666e+000 2.75519657e+000 9.04998899e-001 -1.99772191e+000 2.52548051e+000 -2.79550850e-001 2.24935460e+000 -3.23720884e+000 1.43948674e+000 -1.43738404e-001 2.14141667e-001 1.90374422e+000 -6.48318291e-001 -4.40555632e-001 1.67495656e+000 -1.21303253e-001 2.61465645e+000 3.95782739e-001 1.48943269e+000 -4.29212719e-001 2.21593046e+000 3.06588292e+000 -9.18957964e-002 1.64683378e+000 -2.01091081e-001 1.64259732e-001 1.44116330e+000 -1.27708030e+000 -5.94278574e-001 2.78909802e+000 4.04183298e-001 -6.39155686e-001 -1.11071789e+000 3.91723061e+000 8.88740361e-001 2.70166099e-001 1.68711150e+000 3.41494195e-002 2.62879491e+000 -1.22497030e-001 -8.07257146e-002 -1.23479038e-001 4.16377366e-001 1.49927616e-001 -9.35154483e-002 -2.42750645e-001 7.21661508e-001 6.92346573e-001 1.58245817e-001 -6.56240582e-001 2.31841183e+000 1.18762517e+000 -5.85088849e-001 1.60000634e+000 -4.65205479e+000 8.22968408e-002 -7.74762750e-001 -1.02885211e+000  1.38312590e+000 -8.51595581e-001 1.96487832e+000 2.25681591e+000 1.69004416e+000 3.35943174e+000 1.54889154e+000 3.73949504e+000 1.51178610e+000 -3.16800117e-001 2.04924011e+000 8.30704570e-001 1.65477479e+000 -1.30295336e+000 4.55465412e+000 -1.38525236e+000 1.86339700e+000 1.91202164e+000 9.32169855e-001 -5.06053150e-001 -4.11295146e-002 6.56290650e-001 1.72126293e+000 -2.35860395e+000 1.45457065e+000 -3.16340141e-002 -2.53601027e+000 1.32477188e+000 7.35908270e-001 -3.98547322e-001 1.66871738e+000 3.61932486e-001 -1.42626956e-001 2.08667588e+000 1.06165230e+000 -1.04248536e+000 1.65681597e-002 1.52931511 e+000 2.51952916e-001 -1.65126300e+000 1.22786045e+000 -2.75025427e-001 6.26901507e-001 -1.21067852e-001 1.91679001e+000 -1.16014004e+000 3.37646812e-001 3.64376120e-002 -2.23688388e+000 -1.95473611e+000 1.15201533e+000 -3.13668609e+000 6.84875786e-001 -1.19674683e+000 -1.24230969e+000 -2.26241636e+000 -5.24263144e-001 3.58203673e+000 2.40704998e-001 9.85608280e-001 7.78039932e-001 -7.78025836e-002 5.06118417e-001 -1.16267717e+000 -2.90982664e-001 1.29055846e+000 -4.62567091e-001 3.04141134e-001 3.29742098e+000 -1.57454121e+000 -6.33357108e-001 1.04142857e+000 -8.39529216e-001 1.97137713e+000 1.80137825e+000 -5.80883324e-001 8.53551626e-001 -2.44389606e+000 -1.04786527e+000 6.01655066e-001 1.35108554e+000 1.91133451e+000 -1.55116379e+000 -1.32689738e+000 -3.34982187e-001 -6.04158759e-001 -7.12676525e-001 1.43757379e+000 -3.67825031e-001 1.35954702e+000 5.94304781e-003 -1.79859340e+000 -2.72952080e+000 8.32889140e-001 -1.39144945e+000 -1.17416644e+000 -7.64998674e-001 4.93923604e-001 -1.44545782e+000 1.57880306e+000 1.12437308e+000 1.19925821e+000 7.10867286e-001 4.67830688e-001 6.74548447e-001 1.09623861e+000 2.23827716e-002 -3.27283883e+000 -1.48206964e-001 4.71823335e-001 9.59278345e-001 2.86045492e-001 9.12338674e-001 -2.33875215e-001 1.37641966e-001 2.69536877e+000 2.83514071e+000 -2.39875865e+000 4.55236256e-001 2.48535085e+000 2.72361922e+000 2.54125237e-001 -3.18247676e+000 1.55454111e+000 2.03626370e+000 -6.98356092e-001 -3.20186406e-001 3.07346135e-001 5.24029970e-001 3.58576584e+000 -8.57803285e-001 1.20365873e-001 5.76087952e-001 4.73395920e+000 -4.21760499e-001 2.81904149e+000 1.39611915e-001 -2.31253937e-001 -1.23439991e+000 1.50822794e+000  9.40651536e-001 -7.68359900e-001 2.31346178e+000 -5.87843180e-001 1.24623096e+000 -2.88376665e+000 9.03283894e-001 -1.24732837e-001 2.60412898e-002 2.28092027e+000 -7.80223548e-001 1.42146301e+000 1.28785765e+000 -7.78531671e-001 -1.10662627e+000 3.81465006e+000 1.45004451e+000 -6.86295390e-001 1.31134605e+000 -7.53161073e-001 -2.38736463e+000 9.48513985e-001 2.12668252e+000 2.01590508e-001 2.64392781e+000 -2.63787150e-001 -2.52131611e-001 5.42520463e-001 2.20529389e+000 2.12599683e+000 4.67926890e-001 -1.01627827e+000 -4.38636959e-001 -1.01159893e-001 -6.70790598e-002 1.14079630e+000 2.45170498e+000 -9.42526609e-002 1.41221488e+000 2.27625277e-002 -2.80443698e-001 1.27343559e+000 2.37848088e-001 1.82034159e+000 -8.26801598e-001 -7.50176728e-001 -1.08648551e+000 2.42070699e+000 2.27270246e+000 -1.19815774e-001 -9.50998545e-001 1.68236101e+000 -2.97976494e+000 9.55732107e-001 2.90135597e-003 -2.69438386e+000 3.41107798e+000 1.11490011 e+000 -5.75072527e-001 3.25586051e-001 1.13065195e+000 9.45354939e-001 -3.86584491e-001 -1.10546327e+000 1.85765481e+000 -3.66105199e+000 -1.34011900e+000 -3.60621244e-001 -1.27096891e+000 -7.11437106e-001 2.39505959e+000 1.65358496e+000 1.50321412e+000 3.16486776e-001 4.79341358e-001 1.55303967e+000 3.40338850e+000 4.89003092e-001 2.81465554e+000 -5.24295390e-001 -1.56343615e+000 -1.18851018e+000 3.45909023e+000 5.30569851e-001 1.25108421e+000 -3.48280728e-001 -2.72895396e-001 1.10347736e+000 7.20215678e-001 -2.27794456e+000 -1.01811433e+000 -3.29465437e+000 1.60887623e+000 4.71850276e-001 -2.30937099e+000 -1.09952450e+000 -2.89444745e-001 1.14027429e+000 3.20806056e-001 4.92361672e-002 9.30287540e-001 1.52699798e-001 -1.64544094e+000 1.50056016e+000 -1.19767833e+000 2.22963786e+000 2.25639606e+000 9.56510782e-001 1.05498433e+000 2.60551929e+000 2.47192788e+000 -1.94899297e+000 9.73172724e-001 6.81347847e-001 3.84335667e-002 5.63789785e-001 2.09292388e+000 -2.35492945e+000 -1.60491121e+000 -4.33678776e-001 2.58014894e+000 3.53837281e-001 1.20613635e+000 -1.03824782e+000 -2.70244169e+000 4.62901741e-001 -5.16113460e-001 -2.26867557e+000 7.69846320e-001 -1.06869304e+000 -2.49847388e+000 1.66551188e-001 -2.11130619e-001 1.27242219e+000 2.40089417e+000 -1.50180638e+000 2.83866119e+000 1.66601646e+000 1.21163785e+000 1.51577842e+000  3.16205978e-001 -4.14970517e-001 3.34209830e-001 1.90904868e+000 -6.08527839e-001 2.00294709e+000 -1.42225236e-001 2.44074321e+000 -2.83460450e+000 3.70286083e+000 1.55271268e+000 -1.60934865e-001 -2.05457234e+000 -1.11318493e+000 7.83119619e-001 1.93219930e-001 6.09611750e-001 -8.89916360e-001 -5.18336222e-002 1.62632263e+000 2.05337271e-001 2.36734167e-001 1.08341157e+000 -8.27812910e-001 -1.50358784e+000 7.58716226e-001 -2.18823433e+000 -3.84354234e+000 -7.46470034e-001 1.35967243e+000 -1.51201522e+000 6.61429822e-001 4.33022141e-001 1.31914115e+000 -2.84156471e-001 3.01826668e+000 8.90323818e-002 2.57224154e+000 -3.49536806e-001 1.20527959e+000 -1.64551783e+000 5.97079456e-001 1.46535718e+000 -7.79165745e-001 2.56506968e+000 1.64554584e+000 2.43936133e+000 -1.24797320e+000 1.61060110e-001 2.62977779e-001 6.60554469e-001 -1.77881193e+000 -1.68319678e+000 -3.18312883e-001 -6.25143230e-001 -1.05535913e+000 4.94530886e-001 -5.20972371e-001 -2.16873646e+000 -1.80824995e+000 3.11055136e+000 -3.26495266e+000 -2.33749151e+000 8.06113109e-002 -7.43417799e-001 -1.43765950e+000 1.28656340e+000 9.37331021e-001 1.65003514e+000 2.17272830e+000 5.64124525e-001 3.01361680e+000 1.67168570e+000 2.28130484e+000 1.23492706e+000 6.32456541e-001 1.87143099e+000 -1.76699713e-001 6.86713874e-001 -6.65625334e-002 2.96964973e-001 3.82044941e-001 -3.73141348e-001 1.65588665e+000 -1.86311111e-001 -3.70510578e+000 2.68024445e+000 -2.11436319e+000 2.71787930e+000 -7.34579444e-001 -2.32560158e+000 1.64193237e+000 1.98097467e+000 1.09497666e+000 5.11818349e-001 -3.73851085e+000 4.02708441e-001 2.30055690e+000 -3.67630422e-001 -3.54356766e+000 9.37409978e-003 -1.08068323e+000 2.79761529e+000 6.36268973e-001 -1.53954104e-001 4.92218137e-001 1.00797129e+000 -1.79180205e+000 2.27602243e+000 2.52540797e-001 1.12398112e+000 -2.50184131e+000 1.30374336e+000 1.00840843e+000 7.91947007e-001 1.65224433e-001 -7.91436791e-001 2.79811144e+000 3.53997409e-001 -3.11636615e+000 -2.11673722e-001 4.85001028e-001 2.78055859e+000 -1.47256184e+000 2.09027216e-001 -4.01711315e-001 -2.80359674e+000 7.92990699e-002 -3.77475917e-001 3.30463171e-001 5.17215729e-001 -2.90217614e+000 7.09646821e-001 1.22404337e+000 1.48780787e+000 3.30003649e-001 7.30821073e-001 -3.11456323e+000 4.59069252e-001 -2.16302824e+000  1.05775273e+000 6.04678869e-001 -1.66926908e+000 6.87627673e-001 -1.33886015e+000 -2.73541957e-001 -7.91138187e-002 -4.61171776e-001 -2.02059436e+000 -1.02667737e+000 1.42993629e+000 1.25154507e+000 -5.92566133e-001 1.91667759e+000 1.59074605e+000 1.06317270e+000 3.81611019e-001 -9.54510391e-001 3.52517843e+000 -4.50870705e+000 1.48635769e+000 -2.68166208e+000 -8.87980342e-001 1.16167498e+000 1.40314496e+000 4.81270671e-001 1.04817164e+000 -1.26742125e+000 3.31832290e+000 -2.09496140e+000 -8.00072968e-001 7.72442520e-001 -2.10208988e+000 -8.34703386e-001 -3.58576655e-001 1.82497489e+000 -1.93366349e+000 1.48226321e+000 -7.61157155e-001 4.60760474e-001 -2.66458559e+000 7.54581615e-002 -1.02266443e+000 -3.88942146e+000 -2.46707106e+000 1.35353601 e+000 -3.33155775e+000 -5.69405317e-001 -2.34355664e+000 9.14932430e-001 2.90386176e+000 -9.07698750e-001 2.30322170e+000 -5.37009060e-001 3.27328682e+000 -2.76939601e-001 9.18692470e-001 1.32038641e+000 3.33021188e+000 1.15975368e+000 1.04361510e+000 -2.81390834e+000 -2.22346139e+000 -1.06822634e+000 6.44918501e-001 -4.19576801e-002 -9.63480592e-001 -1.70784652e+000 -4.60445702e-001 -1.37904525e+000 3.25156331e-001 5.62403738e-001 1.78129339e+000 1.33165371e+000 -1.49321783e+000 -1.57136106e+000 9.94577527e-001 4.57476169e-001 2.44409823e+000 -5.55334091e-001 2.25377584e+000 1.19198716e+000 -2.42960095e+000 -1.06827486e+000 3.08647299e+000 1.74202055e-001 -2.06566954e+000 -4.41996425e-001 8.30268204e-001 -3.33665919e+000 -9.09507215e-001 2.08288980e+000 1.35339153e+000 -7.76457965e-001 -8.69336069e-001 2.39780879e+000 2.45157194e+000 -3.29353005e-001 1.56075108e+000 1.19201887e+000 1.26750851e+000 8.99221838e-001 1.90738845e+000 1.81938136e+000 -9.63269055e-001 -7.57165074e-001 1.55659997e+000 -1.92750478e+000 -1.79447556e+000 7.61612654e-001 -1.06798351e+000 -1.52491510e+000 8.09498847e-001 8.15360665e-001 1.45587683e+000 -2.96532065e-001 3.49902302e-001 1.52817357e+000 -3.48740928e-002 8.55526030e-001 -4.30970520e-001 -1.73958808e-001 7.96336412e-001 -6.07606351e-001 -4.60522503e-001 9.14190173e-001 5.89256465e-001 -1.46241927e+000 1.27102017e+000 7.72446930e-001 8.97496998e-001 -2.73194551e-001 -2.78902292e+000 3.15747571e+000 1.36864412e+000 -1.71781838e+000 1.42213500e+000 7.46331871e-001 -9.78936791e-001 -3.39532375e+000  1.66855657e+000 1.07649493e+000 -9.87916648e-001 -9.83610451e-001 1.10034108e+000 4.18890901e-002 3.20622015e+000 3.36752504e-001 2.29340529e+000 -1.53924727e+000 5.37710667e-001 -1.16541696e+000 6.92831993e-001 -1.17176509e+000 2.06588030e+000 1.45995200e+000 -1.28829181e+000 1.19395006e+000 -1.74698621e-001 2.01432729e+000 3.77614737e-001 -1.18296631e-001 -7.40564585e-001 9.01206136e-001 -7.46783733e-001 5.06496355e-002 -1.10556877e+000 2.90351462e+000 1.33163059e+000 -6.31340027e-001 -1.03813303e+000 -1.19116557e+000 -5.86888194e-001 1.15237069e+000 1.42651880e+000 -1.35818005e+000 -2.57812238e+000 3.82667094e-001 6.12372935e-001 1.48428476e+000 1.35794133e-001 7.76758254e-001 -7.03539670e-001 -6.03174090e-001 -6.03082001e-001 1.99687672e+000 -1.18074071e+000 -1.21837628e+000 -2.08009586e-001 -1.05319881e+000 2.63637996e+000 6.30942523e-001 5.15693784e-001 1.76138854e+000 -7.68377483e-001 -4.59355921e-001 6.46186352e-001 -1.86447725e-001 1.31977904e+000 -1.21378517e+000 8.48503828e-001 -1.69902217e+000 6.27434492e-001 -1.20918322e+000 -2.71898794e+000 6.21949911e-001 1.20734927e-004 -2.41284895e+000 -1.65921319e+000 7.24378884e-001 -7.34631896e-001 -2.36548260e-001 4.22343224e-001 1.13314591e-001 -1.44387603e+000 3.21988726e+000 9.78636324e-001 -2.25895897e-001 1.36732399e+000 -1.27787903e-001 -1.14109528e+000 4.24284172e+000 3.68867755e-001 -3.94058943e-001 -1.87558246e+000 -3.67672849e+000 1.35180593e+000 5.44494271e-001 -2.11900043e+000 -8.52134079e-002 -1.05214870e+000 -1.32194981e-001 -1.45356727e+000 3.20472717e-002 6.06840193e-001 -1.63723218e+000 9.62877572e-001 -4.61500555e-001 9.63598549e-001 -8.61029863e-001 -1.06589496e+000 -4.52161729e-001 -8.19720447e-001 1.20012617e+000 1.91720557e+000 2.12556529e+000 -1.88315898e-001 -6.26728415e-001 -6.68102264e-001 1.10652697e+000 2.58852243e-001 2.66154432e+000 -7.34255850e-001 -3.12065445e-002 -2.22151279e+000 -1.37975235e-002 -1.51210678e+000 -9.28165853e-001 4.08833265e+000 -1.14312494e+000 1.29504907e+000 2.74469674e-001 1.71046054e+000 6.18974745e-001 -2.06761813e+000 6.44383669e-001 1.23289537e+000 1.67420253e-001 1.30074763e+000 -1.73213923e+000 4.22435611e-001 -6.98242247e-001 3.73503900e+000 -9.40370619e-001 4.61774921e+000 -1.93580663e+000 -5.14993525e+000 1.67973175e+001 -2.17262897e+001 -9.32018375e+000  -1.15877132e+001 1.62843156e+000
假设计算出一指纹图像的HOG特征x如下，其为一1296维数据：
Figure PCTCN2014094567-appb-000005
Figure PCTCN2014094567-appb-000006
Figure PCTCN2014094567-appb-000007
Figure PCTCN2014094567-appb-000008
Figure PCTCN2014094567-appb-000009
将该指纹图像的HOG特征x输入到最优分类面函数，即代入到如下公式可得：f(x)＝w·x-b＝2.1761881>0。因此可以判断该指纹图像的HOG特征落在f(x)＞0一侧，所以可以判断其为质量好的指纹图像，输出标签值y＝+1。
假设计算出另一指纹图像的HOG特征x如下，其为一1296维数据：
Figure PCTCN2014094567-appb-000010
Figure PCTCN2014094567-appb-000011
Figure PCTCN2014094567-appb-000012
Figure PCTCN2014094567-appb-000013
Figure PCTCN2014094567-appb-000014
将该指纹图像的HOG特征x输入到最优分类面函数，即代入到如下公式可得：f(x)＝w·x-b＝-1.7858665<0。因此可以判断该指纹图像的HOG特征落在f(x)＜0一侧,所以可以判断其为质量差的指纹图像，输出标签值y＝-1。
从而，本发明指纹图像质量的判断方法，通过将HOG特征引入到指纹图像质量判断中，采用机器学习的方式确定最优分类面，根据指纹图像的HOG特征和最优分类面来判断指纹图像质量的好坏。不仅省去人工确定判断阈值的工作，而且具有很好的扩展能力，即这种判断方式可以判定多种类型噪声带来的影响，只需输入需要的样本类型即可完成判断，在大量实验中表现出极佳的效果。本发明的判断方法主要应用于小尺寸指纹采集设备对采集的指纹图像质量好坏进行判断，尤其针对汗渍、泥渍、噪声等造成指纹图像模糊的情况，可以在指纹识别前对指纹图像的质量进行准确的判断，为提高指纹识别率、降低认假率打下了良好的基础。
参见图5、图6，提出本发明指纹图像质量的判断装置一实施例，所述判断装置包括学习模块和判断模块。
学习模块：用于获取指纹图像样本，利用SVM分类器根据指纹图像样本进行学习，获得最优分类面，并将最优分类面的函数和相关参数发送给判断模块。
指纹图像样本包括正负样本，即质量好的样本和质量差的样本，正负样本至少各一个，优选多个。指纹图像样本由人工选择，可以是现场人工采集获取的指纹图像，也可以是从外部获取的现成的指纹图像。
学习模块首先计算指纹图像样本的HOG特征，然后将指纹图像样本的HOG特征输入SVM分类器中进行训练学习，从而获得最优分类面。
计算HOG特征时，学习模块首先计算指纹图像样本中每一像素位置的梯度方向值；接着将指纹图像样本分成多个单元格，根据梯度方向值为每个单元格构建梯度方向直方图；然后将单元格组合成块，在块内归一化梯度方向直方图；最后将指纹图像样本中所有块的梯度方向直方图结合起来形成HOG特征。
学习模块将多个指纹图像样本的HOG特征输入SVM，SVM则根据多个指纹图像样本的HOG特征进行训练学习，获得最优分类面。该最优分类面的函数为f(x)＝w·x-b，其中w为最优分类面的支持向量，b为常数项，x为输入的图像的HOG特征，w和b为最优分类面的参数。
判断模块：用于获取待判断的指纹图像，并计算该指纹图像的HOG特征，根据HOG特征和最优分类面判断指纹图像的质量，并输出判断结果。
判断模块包括处理单元和判别单元，其中：
处理单元：用于获取待判断的指纹图像，计算指纹图像的HOG特征，将该HOG特征输入最优分类面的函数中进行计算，并向判别单元发送判断结果。
具体的，处理单元首先计算指纹图像中每一像素位置的梯度方向值；接着将指纹图像分成多个单元格，根据梯度方向值为每个单元格构建梯度方向直方图；然后将单元格组合成块，在块内归一化梯度方向直方图，将指纹图像中所有块的梯度方向直方图结合起来形成HOG特征。最后，处理单元将指纹图像的HOG特征输入最优分类面的函数f(x)＝w*x-b中进行计算，其中w为最优分类面的支持向量，b为常数项，x为指纹图像的HOG特征。
判别单元：用于根据处理单元的计算结果进行判别，若计算结果为f(x)＞0，则判定指纹图像的质量好，输出标签值y＝+1；若计算结果为f(x)＜0，则判定指纹图像的质量差，输出标签值y＝-1。
据此，本发明指纹图像质量的判断装置，通过引入HOG特征并采用SVM 监督学习模式成功地区分出指纹图像质量，尤其针对汗渍、泥渍、噪声等造成指纹图像模糊的情况，可以在指纹识别前对指纹图像的质量进行准确的判断，为提高识别率、降低认假率打下良好的基础，尤其适用于小尺寸指纹采集设备。本发明的判断装置，不仅省去人工确定判断阈值的工作，而且具有很好的扩展能力，即这种判断方式可以判定多种类型噪声带来的影响，只需输入需要的样本类型即可完成判断，在大量实验中表现出极佳的效果。
需要说明的是：上述实施例提供的指纹图像质量的判断装置在进行指纹图像质量的判断时，仅以上述各功能模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能模块完成。另外，上述实施例提供的指纹图像质量的判断装置与指纹图像质量的判断方法实施例属于同一构思，其具体实现过程详见方法实施例，且方法实施例中的技术特征在装置实施例中均对应适用，这里不再赘述。
本领域普通技术人员可以理解,实现上述实施例方法中的全部或部分步骤可以通过程序来控制相关的硬件完成，所述的程序可以存储于一计算机可读取存储介质中，所述的存储介质可以是ROM/RAM、磁盘、光盘等。
应当理解的是，以上仅为本发明的优选实施例，不能因此限制本发明的专利范围，凡是利用本发明说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本发明的专利保护范围内。
工业实用性
本发明所提供的一种指纹图像质量的判断方法，通过SVM分类器对指纹图像样本进行学习获得最优分类面，并将HOG特征引入到指纹图像质量判断中，根据HOG特征和最优分类面自动判断指纹图像的质量。不仅省去了人工确定判断阈值的工作，而且具有很好的扩展能力，即这种判断方式可以判定多种类型噪声带来的影响，只需输入需要的样本类型即可完成判断，在大量实验中表现出极佳的效果。本发明的判断方法主要适用于小尺寸指纹采集设备对采集的指纹图像质量好坏的判断，尤其针对汗渍、泥渍、噪声等造成指纹图像模糊的情形，可以在指纹识别前对指纹图像的质量进行准确的判断，为提高指纹识别率、降低认假率打下了良好的基础。

Claims (8)

  1. 一种指纹图像质量的判断方法，包括步骤：
    获取指纹图像样本；
    利用SVM支持向量机分类器根据所述指纹图像样本进行学习，获得最优分类面；
    获取待判断的指纹图像，并计算所述指纹图像的HOG梯度方向直方图特征；
    根据所述HOG特征和最优分类面判断所述指纹图像的质量。
  2. 根据权利要求1所述的指纹图像质量的判断方法，其中，所述利用SVM支持向量机分类器根据所述指纹图像样本进行学习获得最优分类面包括：
    计算所述指纹图像样本的HOG特征；
    将所述指纹图像样本的HOG特征输入所述SVM分类器中进行训练学习，获得最优分类面。
  3. 根据权利要求1所述的指纹图像质量的判断方法，其中，所述计算所述指纹图像的HOG梯度方向直方图特征包括：
    计算所述指纹图像中每一像素位置的梯度方向值；
    将所述指纹图像分成多个单元格，根据所述梯度方向值为每个单元格构建梯度方向直方图；
    将所述单元格组合成块，在所述块内归一化所述梯度方向直方图，将所述指纹图像中所有块的梯度方向直方图结合起来形成HOG特征。
  4. 根据权利要求1所述的指纹图像质量的判断方法，其中，所述根据所述HOG特征和最优分类面判断所述指纹图像的质量包括：
    将所述指纹图像的HOG特征输入所述最优分类面的函数f(x)＝w*x-b中进行计算，其中w为最优分类面的支持向量，b为常数项，x为指纹图像的HOG特征；
    若计算结果为f(x)＞0，则判定所述指纹图像的质量好；
    若计算结果为f(x)＜0，则判定所述指纹图像的质量差。
  5. 一种指纹图像质量的判断装置，包括学习模块和判断模块，其中：
    学习模块，设置为获取指纹图像样本，利用SVM分类器根据所述指纹图像样本进行学习，获得最优分类面；
    判断模块，设置为获取待判断的指纹图像，并计算所述指纹图像的HOG梯度方向直方图特征，根据所述HOG特征和最优分类面判断所述指纹图像的质量。
  6. 根据权利要求5所述的指纹图像质量的判断装置，其中，所述学习模块设置为：计算所述指纹图像样本的HOG特征，将所述指纹图像样本的HOG特征输入所述SVM分类器中进行训练学习，获得最优分类面。
  7. 根据权利要求5所述的指纹图像质量的判断装置，其中，所述判断模块包括处理单元，所述处理单元设置为：计算所述指纹图像中每一像素位置的梯度方向值；将所述指纹图像分成多个单元格，根据所述梯度方向值为每个单元格构建梯度方向直方图；将所述单元格组合成块，在所述块内归一化所述梯度方向直方图，将所述指纹图像中所有块的梯度方向直方图结合起来形成HOG特征。
  8. 根据权利要求5所述的指纹图像质量的判断装置，其中，所述判断模块包括处理单元和判别单元，其中：
    处理单元，设置为将所述指纹图像的HOG特征输入所述最优分类面的函数f(x)＝w*x-b中进行计算，其中w为最优分类面的支持向量，b为常数项，x为指纹图像的HOG特征；
    判别单元，设置为根据处理单元的计算结果进行判别，若计算结果为f(x)＞0，则判定所述指纹图像的质量好；若计算结果为f(x)＜0，则判定所述指纹图像的质量差。
PCT/CN2014/094567 2014-09-28 2014-12-22 指纹图像质量的判断方法和装置 WO2016045215A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN2014105102270 2014-09-28
CN201410510227.0A CN104268529A (zh) 2014-09-28 2014-09-28 指纹图像质量的判断方法和装置

Publications (1)

Publication Number Publication Date
WO2016045215A1 true WO2016045215A1 (zh) 2016-03-31

Family

ID=52160049

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2014/094567 WO2016045215A1 (zh) 2014-09-28 2014-12-22 指纹图像质量的判断方法和装置

Country Status (2)

Country Link
CN (1) CN104268529A (zh)
WO (1) WO2016045215A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110838116A (zh) * 2019-11-14 2020-02-25 上海联影医疗科技有限公司 医学图像采集方法、装置、设备和计算机可读存储介质

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106408552B (zh) * 2015-07-27 2019-07-12 辽宁省公安厅 一种基于秩和比方法的多指位指纹图像质量综合评价方法
CN105718848B (zh) * 2015-10-21 2020-08-07 深圳芯启航科技有限公司 一种指纹图像的质量评估方法及装置
CN105528591B (zh) * 2016-01-14 2019-04-16 电子科技大学 基于多象限编码的活体指纹识别方法
CN106355145A (zh) * 2016-08-26 2017-01-25 广东欧珀移动通信有限公司 指纹识别方法、装置及移动终端
CN106529545B (zh) * 2016-09-26 2019-11-26 北京林业大学 一种基于图像特征描述的散斑图像质量识别方法及系统
CN108985351B (zh) * 2018-06-27 2021-11-26 北京中安未来科技有限公司 一种基于梯度方向稀疏特征信息识别模糊图像的方法和装置、计算设备及存储介质
CN110472518B (zh) * 2019-07-24 2022-05-17 杭州晟元数据安全技术股份有限公司 一种基于全卷积网络的指纹图像质量判断方法
CN111428748B (zh) * 2020-02-20 2023-06-27 重庆大学 一种基于hog特征和svm的红外图像绝缘子识别检测方法
CN112699863B (zh) * 2021-03-25 2022-05-17 深圳阜时科技有限公司 指纹增强方法、计算机可读存储介质及电子设备
CN115841685B (zh) * 2023-02-15 2023-05-12 南京信息工程大学 一种基于复合像元梯度的伪造指纹检测系统及方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7853072B2 (en) * 2006-07-20 2010-12-14 Sarnoff Corporation System and method for detecting still objects in images
CN102169533A (zh) * 2011-05-11 2011-08-31 华南理工大学 一种商用网页恶意篡改检测方法
CN102799669A (zh) * 2012-07-17 2012-11-28 杭州淘淘搜科技有限公司 一种商品图像视觉质量的自动分级方法
CN102915446A (zh) * 2012-09-20 2013-02-06 复旦大学 基于svm机器学习的植物病虫害检测方法
CN102982350A (zh) * 2012-11-13 2013-03-20 上海交通大学 一种基于颜色和梯度直方图的台标检测方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN100592323C (zh) * 2008-07-01 2010-02-24 山东大学 面向图像质量的指纹识别方法
CN102831409B (zh) * 2012-08-30 2016-09-28 苏州大学 基于粒子滤波的运动行人视频自动跟踪方法及系统
US20140226024A1 (en) * 2013-02-08 2014-08-14 Kutta Technologies, Inc. Camera control in presence of latency

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7853072B2 (en) * 2006-07-20 2010-12-14 Sarnoff Corporation System and method for detecting still objects in images
CN102169533A (zh) * 2011-05-11 2011-08-31 华南理工大学 一种商用网页恶意篡改检测方法
CN102799669A (zh) * 2012-07-17 2012-11-28 杭州淘淘搜科技有限公司 一种商品图像视觉质量的自动分级方法
CN102915446A (zh) * 2012-09-20 2013-02-06 复旦大学 基于svm机器学习的植物病虫害检测方法
CN102982350A (zh) * 2012-11-13 2013-03-20 上海交通大学 一种基于颜色和梯度直方图的台标检测方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110838116A (zh) * 2019-11-14 2020-02-25 上海联影医疗科技有限公司 医学图像采集方法、装置、设备和计算机可读存储介质
CN110838116B (zh) * 2019-11-14 2023-01-03 上海联影医疗科技股份有限公司 医学图像采集方法、装置、设备和计算机可读存储介质

Also Published As

Publication number Publication date
CN104268529A (zh) 2015-01-07

Similar Documents

Publication Publication Date Title
WO2016045215A1 (zh) 指纹图像质量的判断方法和装置
US9633269B2 (en) Image-based liveness detection for ultrasonic fingerprints
EP3091479B1 (en) Method and apparatus for fingerprint identification
CN111899246B (zh) 玻片数字化信息质量检测方法、装置、设备及介质
CN104036284A (zh) 基于Adaboost算法的多尺度行人检测方法
Laga et al. Image-based plant stornata phenotyping
CN105279772A (zh) 一种红外序列图像的可跟踪性判别方法
CN108073940B (zh) 一种非结构化环境中的3d目标实例物体检测的方法
Kruthi et al. Offline signature verification using support vector machine
CN105488486A (zh) 防止照片攻击的人脸识别方法及装置
Mammeri et al. North-American speed limit sign detection and recognition for smart cars
CN110619280B (zh) 一种基于深度联合判别学习的车辆重识别方法及装置
US11468572B2 (en) Image processing device, image recognition device, image processing program, and image recognition program
CN105809092A (zh) 人群目标检测方法及其装置
CN116703895B (zh) 基于生成对抗网络的小样本3d视觉检测方法及其系统
Cao et al. Power line detection based on symmetric partial derivative distribution prior
CN112396638A (zh) 一种图像处理方法、终端和计算机可读存储介质
WO2018143278A1 (ja) 画像処理装置、画像認識装置、画像処理プログラム、及び画像認識プログラム
Tian et al. Robust traffic sign detection in complex road environments
Yang et al. Qualifying fingerprint samples captured by smartphone cameras
CN110222666A (zh) 一种签名鉴伪方法和系统
CN114092743B (zh) 敏感图片的合规性检测方法、装置、存储介质及设备
Tao et al. Pedestrian Identification and Tracking within Adaptive Collaboration Edge Computing
CN113610790B (zh) 一种基于图像识别的气体扩散层纤维测量方法
Pan et al. The Crime Scene Tools Identification Algorithm Based on GVF-Harris-SIFT and KNN.

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: 14902857

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: 14902857

Country of ref document: EP

Kind code of ref document: A1