CN106203262A

CN106203262A - A kind of ocular form sorting technique based on eyelid curve similarity Yu ocular form index

Info

Publication number: CN106203262A
Application number: CN201610486235.5A
Authority: CN
Inventors: 孙劲光; 荣文钊; 王伟
Original assignee: Liaoning Technical University
Current assignee: Liaoning Technical University
Priority date: 2016-06-27
Filing date: 2016-06-27
Publication date: 2016-12-07

Abstract

Aiming at the problem of human eye type classification, using different shapes of eyelid curves, a human eye classification method based on the combination of eyelid contour curve similarity and eye shape index is proposed. Firstly, the contour-based shape description method is used to add sampling points on the basis of eye feature points. Then, according to the upper and lower eyelid sampling points and the inner and outer corner points, the upper and lower eyelid curve equations are obtained through least squares fitting. Then, the slope and eye shape index of the upper and lower eyelid sampling points are calculated through the eyelid curve equation. Finally, the normalized cross-correlation coefficient is used to describe the slope similarity of the upper and lower eyelid sampling points. Eye type classification is realized according to the similarity of eye type sample curves and eye type index. This method is used to classify four eye types: standard eye, round eye, squint eye and slender eye. Experiments show that the proposed method is easy to implement and has a good effect on eye type classification. This method can be applied in all occasions such as kinship recognition, face recognition, etc. that need to recognize specific human eye types.

Description

An Eye Type Classification Method Based on Eyelid Curve Similarity and Eye Type Index

技术领域technical field

本发明属于图像识别技术，特别涉及一种基于眼睑曲线相似度与眼型指数的眼型分类方法。The invention belongs to image recognition technology, in particular to an eye shape classification method based on eyelid curve similarity and eye shape index.

背景技术Background technique

文献[1]提出利用Active Shape Model进行眼型分类，这种方法只对左眼、右眼进行了区分，并没有针对不同眼睛形状进行眼型分类。并且该方法中仅仅是把ASM特征点进行直线连线，这种眼睛形状描述方式会损失眼睑曲线弧度，而眼睑曲线是对眼型分类的关键所在。目前，没有针对具体眼型进行分类的方法，但受脸型分类方法^[2-4]和眼部轮廓提取方法^[5-6]的启发。根据不同眼睑的形状，把眼睛分为标准眼、圆眼、眯缝眼、细长眼四种类型，由于眼睛形状的“一般性”，无法使用刚性形状模型来描述。Literature [1] proposes to use Active Shape Model to classify eye types. This method only distinguishes left and right eyes, and does not classify eye types for different eye shapes. Moreover, this method only connects the ASM feature points with straight lines. This eye shape description method will lose the radian of the eyelid curve, and the eyelid curve is the key to eye shape classification. Currently, there is no method for classifying specific eye shapes, but it is inspired by face shape classification methods ^[2-4] and eye contour extraction methods ^[5-6] . According to the shape of different eyelids, the eyes are divided into four types: standard eyes, round eyes, squint eyes, and slender eyes. Due to the "generality" of eye shapes, rigid shape models cannot be used to describe them.

参考文献：references:

[1]S.Bhat and M.Savvides,Evaluating Active Shape Models for Eye-ShapeClassification,2008 IEEE International Conference on Acoustics,Speech andSignal Processing,Las Vegas,NV,2008,pp.5228-5231.[1] S.Bhat and M.Savvides, Evaluating Active Shape Models for Eye-Shape Classification, 2008 IEEE International Conference on Acoustics, Speech and Signal Processing, Las Vegas, NV, 2008, pp.5228-5231.

[2]张倩,丁友东,蓝建梁,涂意.基于ASM和K近邻算法的人脸脸型分类[J].计算机工程,2011,11:212-214+217.[2] Zhang Qian, Ding Youdong, Lan Jianliang, Tu Yi. Classification of Face Shapes Based on ASM and K Nearest Neighbor Algorithm [J]. Computer Engineering, 2011, 11: 212-214+217.

[3]王俊艳,苏光大.基于下颌轮廓线的人脸分类方法[J].红外与激光工程,2004,02:159-163.[3] Wang Junyan, Su Guangda. Face Classification Method Based on Mandibular Contour [J]. Infrared and Laser Engineering, 2004, 02:159-163.

[4]赵薇,汪增福.用于大库人脸识别的脸型分类研究[J].电子技术,2009,11:77-79+68.[4] Zhao Wei, Wang Zengfu. Research on Face Classification for Face Recognition in Daku [J]. Electronic Technology, 2009, 11:77-79+68.

[5]刘伟锋.人脸表情识别研究[D].中国科学技术大学,2007.[5] Liu Weifeng. Research on Facial Expression Recognition [D]. University of Science and Technology of China, 2007.

[6]魏博,李战明.傅里叶变换在人眼轮廓拟合和人眼几何参数计算中的应用[J].电子设计工程,2015,20:177-180.[6] Wei Bo, Li Zhanming. Application of Fourier Transform in Human Eye Contour Fitting and Human Eye Geometric Parameter Calculation [J]. Electronic Design Engineering, 2015, 20:177-180.

发明内容Contents of the invention

为了解决上述技术问题，本发明提供一种使用眼睑曲线相似度与眼型指数相结合的眼型分类方法，本方法重点解决在已知眼部特征点的基础上如何得到上、下眼睑轮廓曲线、如何描述眼睑轮廓曲线相似度以及如何利用眼睑轮廓曲线相似度与眼型指数相结合的方法将眼型进行分类这三个问题。In order to solve the above-mentioned technical problems, the present invention provides an eye type classification method that combines eyelid curve similarity and eye shape index. This method focuses on how to obtain the upper and lower eyelid contour curves on the basis of known eye feature points 1. How to describe the similarity of eyelid contour curves and how to classify eye types by combining the similarity of eyelid contour curves with the eye shape index.

解决上述技术问题所采用的技术方案是：The technical scheme adopted to solve the above-mentioned technical problems is:

一种基于眼睑曲线相似度与眼型指数的眼型分类方法，其特征在于，通过计算待识别眼型与各类型标准眼型之间眼睑轮廓曲线相似度和眼型指数差达到眼型分类的目的，所述眼型分类方法的步骤如下：An eye type classification method based on eyelid curve similarity and eye type index, characterized in that the eye type classification is achieved by calculating the eyelid contour curve similarity and eye type index difference between the eye type to be recognized and various types of standard eye types Purpose, the step of described eye type classification method is as follows:

令Sample_ij为标准样本，i＝1,2,3,4,四类眼型，j＝1,2,…,N,每类眼型有N个标准样本；Unknow为待识别眼型；K-Unknow为待识别眼型的类型；Let Sample _ij be a standard sample, i=1,2,3,4, four eye types, j=1,2,...,N, each eye type has N standard samples; Unknow is the eye type to be identified; K -Unknown is the type of eye type to be recognized;

1)从Sample_ij中提取特征点向量f_ij(x₀,y₀,…,x₁₂,y₁₂)，由所有特征点向量构成特征点集合F；1) Extract feature point vector f _ij (x ₀ ,y ₀ ,…,x ₁₂ ,y ₁₂ ) from Sample _ij , and form feature point set F from all feature point vectors;

2)由每一个特征点向量f_ij，计算得到一个采样点向量p_ij(x₀,y₀,…,x₄₇,y₄₇)，由所有采样点构成采样点集合P；2) A sampling point vector p _ij (x ₀ ,y ₀ ,…,x ₄₇ ,y ₄₇ ) is calculated from each feature point vector f _ij , and a sampling point set P is composed of all sampling points;

3)把每一个采样点向量p_ij分为上下眼睑(x₀，y₀，…，x_m，y_m)，(x₄₇，y₄₇，…，x_m+1，y_m+1)两部分，如果内外眼角点不包含在(x₀，y₀，…，x_m，y_m)，(x_m+1，y_m+1，…，x₄₇，y₄₇)中，则把内外眼角点插入，构成(x_内，y_内，x₀，y₀，…，x_m，y_m，x_外，y_外)，(x_内，y_内，x₄₇，y₄₇，…，x_m+1，y_m+1，x_外，y_外)；3) Divide each sampling point vector p _ij into upper and lower eyelids (x ₀ , y ₀ , ..., x _m , y _m ), (x ₄₇ , y ₄₇ , ..., x _m+1 , y _m+1 ) two part, if the inner and outer canthus points are not included in (x ₀ , y ₀ , ..., x _m , y _m ), (x _m+1 , y _m+1 , ..., x ₄₇ , y ₄₇ ), then put the inner and outer canthus Point insertion, constitute (x _inside , y _inside , x ₀ , y ₀ , ..., x _m , y _m , x _outside , y _outside ), (x _inside , y _inside , x ₄₇ , y ₄₇ , ..., x _{m+ 1} , y _m+1 , x _outside , y _outside );

4)使用(x_内,y_内,x₀,y₀,…,x_m,y_m,x_外,y_外)、(x_内,y_内,x₄₇,y₄₇,…,x_m+1,y_m+1,x_外,y_外)进行最小二乘法函数拟合得到上、下眼睑的曲线方程系数向量a_ij(c₀,…,c₉)，c₀,…,c₄为上眼睑曲线方程系数，c₅,…,c₉为下眼睑曲线方程系数，由曲线方程系数向量构成系数集合A；4) Use (x _inside ,y _inside ,x ₀ ,y ₀ ,…,x _m ,y _m ,x _outside ,y _outside ), (x _inside ,y _inside ,x ₄₇ ,y ₄₇ ,…,x _m+1 ,y _m+1 , _outside x _,outside y) to get the curve equation coefficient vector a _ij (c ₀ ,…,c ₉ ) of the upper and lower eyelids by least square function fitting, and c ₀ ,…,c ₄ are upper The coefficients of the eyelid curve equation, c ₅ ,..., c ₉ are the coefficients of the lower eyelid curve equation, and the coefficient set A is composed of the curve equation coefficient vector;

5)使用采样点向量p_ij与系数向量a_ij计算曲线采样点切线斜率构成斜率向量s_ij(n₀,…,n₄₇)，由所有斜率向量构成斜率集合S；5) Use the sampling point vector p _ij and the coefficient vector a _ij to calculate the tangent slope of the curve sampling point to form a slope vector s _ij (n ₀ ,...,n ₄₇ ), and all slope vectors form a slope set S;

6)使用采样点向量p_ij与系数向量a_ij计算眼型指数e_ij，由所有眼型指数构成眼型指数集合E；6) Use the sampling point vector p _ij and the coefficient vector a _ij to calculate the eye shape index e _ij , and form the eye shape index set E from all the eye shape indexes;

7)计算Unknow的斜率向量s_unknow(n₀,…,n₄₇)、眼型指数e_unknow；7) Calculate _Unknow 's slope vector s unknown (n ₀ ,...,n ₄₇ ), eye shape index e _unknown ;

8)计算s_unknow与斜率集合S列向量相似度C(c₁,c₂,c₃,c₄)，计算e_unknow与眼型指数集合E列向量的差和D(d₁,d₂,d₃,d₄)；8) Calculate the similarity C(c ₁ ,c ₂ ,c ₃ ,c ₄ ) between _{sun unknown} and the slope set S column vector, calculate the difference between e _unknown and the eye shape index set E column vector and D(d ₁ ,d ₂ , d ₃ , d ₄ );

9)由相似度向量C与差和向量D，判别Unknow的类别K-Unknow，输出K-Unknow。9) Based on the similarity vector C and the difference sum vector D, the category K-Unknow of Unknow is judged, and K-Unknow is output.

1上下眼睑的轮廓曲线方程的拟合1 Fitting of contour curve equations of upper and lower eyelids

1.1上下眼睑曲线采样点的求取1.1 Obtaining the sampling points of the upper and lower eyelid curves

将眼部特征点F_i(i＝0…11)，(F₁₂为瞳孔点，F₀为内眼角点，F₆为外眼角点，F₁—F₅为上眼睑点，F₇—F₁₁为下眼睑点)依次连线得到眼睛轮廓多边形。以瞳孔点F₁₂为圆心点，以过瞳孔点平行X轴直线为0°线，进行等圆心角采样，将圆心等分48等分，等分线与眼睛轮廓多边形相交得采样点P_i(i＝0，…，47)，即四种眼型采样点分布情况如图1所示。通过内眼角点F₀与外眼角点F₆连线，把48个采样点P_i(i＝0，…，47)，分为上眼睑采样点P_i(i=0，…，m)，下眼睑采样点P_i(i=m+1，…，47)。The eye feature points F _i (i=0...11), (F ₁₂ is the pupil point, F ₀ is the inner corner point, F ₆ is the outer corner point, F ₁ -F ₅ is the upper eyelid point, F ₇ -F ₁₁ is the lower eyelid point) and connects successively to obtain the eye contour polygon. Take the pupil point _F12 as the center point, and take the straight line parallel to the X-axis passing through the pupil point as the 0° line to sample the _isocentral angle, divide the center of the circle into 48 equal parts, and obtain the sampling point Pi( i=0,...,47), that is, the distribution of the four eye-type sampling points is shown in Figure 1. Through the connection between the inner corner point F ₀ and the outer corner point F ₆ , the 48 sampling points P _i (i=0,...,47) are divided into upper eyelid sampling points P _i (i=0,...,m), Lower eyelid sampling point P _i (i=m+1, . . . , 47).

1.2眼睑曲线拟合的基函数选取1.2 Selection of basis functions for eyelid curve fitting

最小二乘算法是以误差的平方和最小为准则，根据观测数据估计线性模型中未知参数的一种基本参数估计方法。它通过最小化误差的平方和寻找数据的最佳函数匹配，利用最小二乘法可以进行曲线拟合。上、下眼睑曲线拟合的基函数为只要满足n<m且n<48-m，可以得到唯一眼睑曲线方程。由1.3可知上、下眼睑采样点，内眼角点F₀，外眼角点F₆满足n<m且n<48-m，所以可以使用最小二乘进行曲线拟合，分别将上眼睑采样点P_i(i＝0…m)，内眼角点F₀，外眼角点F₆；下眼睑采样点P_i(i＝m+1，…，47)，内眼角点F₀，外眼角点F₆进行曲线拟合(如果内外眼角点与采样点重复则取其一)分别得到最接近采样点的上、下眼睑曲线方程公式(1)。The least squares algorithm is a basic parameter estimation method for estimating the unknown parameters in the linear model based on the observed data based on the criterion of the minimum sum of the squares of the errors. It finds the best function match for the data by minimizing the sum of the squares of the errors, and curve fitting can be performed using the least squares method. The basis function of upper and lower eyelid curve fitting is As long as n<m and n<48-m are satisfied, the unique eyelid curve equation can be obtained. From 1.3, it can be seen that the upper and lower eyelid sampling points, the inner corner point F ₀ , and the outer corner point F ₆ satisfy n<m and n<48-m, so the least squares can be used for curve fitting, and the upper eyelid sampling point P _i (i=0...m), inner corner point F ₀ , outer corner point F ₆ ; lower eyelid sampling point P _i (i=m+1,...,47), inner corner point F ₀ , outer corner point F ₆ Carry out curve fitting (if the inner and outer corners of the eyes overlap with the sampling points, choose one) to obtain the upper and lower eyelid curve equations (1) closest to the sampling points respectively.

其中，当*为up时为上眼睑轮廓曲线方程，当*为down时为下眼睑轮廓曲线方程，n为拟合次数，为曲线方程系数，为基函数。Among them, when * is up, it is the upper eyelid contour curve equation, when * is down, it is the lower eyelid contour curve equation, n is the number of fittings, is the coefficient of the curve equation, is the base function.

1.3拟合次数的选定1.3 Selection of fitting times

由于眼角点对于上、下眼睑曲线的确定有着至关重要的作用，通过上、下眼睑的采样点拟合到的眼睑曲线必须经过内眼角点F₀、外眼角点F₆，取基通过大量实验比较，只有在拟合次数达到4次以上时，上、下眼睑的内、外眼角点位置拟合效果才能达到要求如图2所示，所以取拟合次数n＝4。Since the corners of the eyes play a crucial role in determining the curves of the upper and lower eyelids, the eyelid curves fitted by the sampling points of the upper and lower eyelids must pass through the inner corner of the eye F ₀ and the outer corner of the eye F ₆ . Through a large number of experimental comparisons, only when the number of fitting times reaches more than 4 times, the fitting effect of the inner and outer corners of the upper and lower eyelids can meet the requirements, as shown in Figure 2, so the number of fitting times n=4.

1.4上下眼睑的轮廓曲线方程1.4 Contour curve equation of the upper and lower eyelids

取基4次拟合眼部轮廓曲线方程为公式(4)，求的法方程为(3)。Take the base The 4th degree fitting eye contour curve equation is formula (4), find The normal equation is (3).

分别将圆眼、标准眼、细长眼、眯缝眼的内、外眼角点与上眼睑采样点P_i(i＝0，…，m)，下眼睑采样点P_i(i＝m+1…47)，带入法方程即可求出相应圆眼上、下眼睑轮廓曲线方程为公式(5)，标准眼上、下眼睑轮廓曲线方程为公式(5)，细长眼上、下眼睑轮廓曲线方程为公式(5)，眯缝眼上、下眼睑轮廓曲线方程为公式(5)。Respectively compare the inner and outer corner points of round eyes, standard eyes, slender eyes, and squint eyes with upper eyelid sampling points P _i (i=0,...,m), lower eyelid sampling points P _i (i=m+1... 47), put into the normal equation to get the corresponding The contour curve equation of upper and lower eyelids for round eyes is formula (5), the contour curve equation of upper and lower eyelids for standard eyes is formula (5), and the contour curve equation of upper and lower eyelids for slender eyes is formula (5). The equation of the lower eyelid contour curve is formula (5).

$[\begin{matrix} m m & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{22} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{33} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{44} \\ {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{22} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{33} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{44} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{55} \\ {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{22} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{33} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{44} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{55} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{66} \\ {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{33} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{44} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{55} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{66} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{77} \\ {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{44} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{55} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{66} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{77} & {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{88} \end{matrix}] [\begin{matrix} {a a}_{00} \\ {a a}_{11} \\ {a a}_{22} \\ {a a}_{33} \\ {a a}_{44} \end{matrix}] = = [\begin{matrix} {Σ Σ}_{i i = = 11}^{m m} {y the y}_{i i} \\ {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{11} {y the y}_{i i} \\ {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{22} {y the y}_{i i} \\ {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{33} {y the y}_{i i} \\ {Σ Σ}_{i i = = 11}^{m m} {x x}_{i i}^{44} {y the y}_{i i} \end{matrix}] - - - - - - ((33))$

其中m为采样点个数where m is the number of sampling points

${S S}^{* *} ((x x)) = = {Σ Σ}_{k k = = 00}^{44} {a a}_{k k}^{* *} {x x}^{k k} - - - - - - ((44))$

${S S}_{t t}^{* *} ((x x)) = = {Σ Σ}_{k k = = 00}^{44} {a a}_{t t k k}^{* *} {x x}^{k k},, t t = = 11,, ... ...,, 44 - - - - - - ((55))$

其中t＝1代表圆眼、t＝2代表标准眼、t＝3代表细长眼、t＝4代表眯缝眼。Among them, t=1 represents round eyes, t=2 represents standard eyes, t=3 represents slender eyes, and t=4 represents squint eyes.

2眼睑曲线相似度判别方法2 Discrimination method of eyelid curve similarity

不同类别眼型在相同位置采样点的眼睑切线坡度是不同的，而同类别眼型在相同位置采样点的眼睑切线坡度是相似的，所以使用眼睑采样点的斜率特征来描述眼睑曲线之间的相似度。并且斜率的几何特征明显，不受缩放和位移的影响，可以得到良好的分类效果。对拟合得到的上、下眼睑的曲线方程求导后，通过公式(6)可以得到上、下眼睑采样点的曲线切线斜率值。The eyelid tangent slopes of different types of eye types at the same location sampling points are different, while the eyelid tangent slopes of the same type of eye types at the same location sampling points are similar, so the slope characteristics of eyelid sampling points are used to describe the relationship between eyelid curves similarity. Moreover, the geometric characteristics of the slope are obvious, and it is not affected by scaling and displacement, and a good classification effect can be obtained. After deriving the fitted curve equations of the upper and lower eyelids, the slope values of the curve tangents at the sampling points of the upper and lower eyelids can be obtained by formula (6).

通过公式(7)计算得ρ为两个上眼睑或两个下眼睑曲线的相似度。当ρ越大时，两个眼型曲线切线斜率的相似度越大，两个眼型为同类型的程度越高，反之表示两个眼型越不相同，不能为同类型眼型。|p|＝0.95为区分不同眼型的相似度阈值效果比较好。Calculated by formula (7), ρ is the similarity of two upper eyelids or two lower eyelid curves. When ρ is larger, the similarity between the slopes of the tangent lines of the two eye shapes is greater, and the degree of the two eye shapes being of the same type is higher. On the contrary, it means that the two eye shapes are more different and cannot be of the same type. |p|=0.95 means that the similarity threshold for distinguishing different eye types is better.

${S S}^{* *' '} ((x x)) = = \frac{\partial \partial {S S}^{* *}}{\partial \partial x x} = = {Σ Σ}_{k k = = 11}^{44} k k \cdot &Center Dot; {a a}_{k k} {x x}^{k k - - 11} - - - - - - ((66))$

$ρ ρ = = \frac{{Σ Σ}_{i i = = 11}^{N N} X x ((i i)) Y Y ((i i))}{\sqrt{{Σ Σ}_{i i = = 11}^{N N} X x {((i i))}^{22} {Σ Σ}_{i i = = 11}^{N N} Y Y {((i i))}^{22}}} - - - - - - ((77))$

其中分别为需要比较的两个上眼睑或者两个下眼睑采样点，N为采样点个数。in are the two upper eyelids or two lower eyelid sampling points to be compared, and N is the number of sampling points.

3眼型指数3 eye type index

通过测量法对不同类型的眼型进行测量发现，不同类型眼型的眼型指数有着明显的区别。眼型指数E_i，e_h、e_w如图3所示，得出不同眼型的眼型指数大致范围，见表1，证明眼型指数可以作为判别眼型特征。By measuring different types of eye shapes, it is found that there are obvious differences in the eye shape indexes of different types of eye shapes. The eye shape indexes E _i , e _h , and e _w are shown in Figure 3, and the approximate ranges of eye shape indexes for different eye types are obtained, as shown in Table 1, which proves that the eye shape indexes can be used as the characteristics for distinguishing eye shapes.

${E E.}_{i i} = = \frac{| | {e e}_{h h} | |}{| | {e e}_{w w} | |},, i i = = 11,, 22,, 33,, 44 - - - - - - ((88))$

其中，e_w＝F₀-F₆ in, e _w =F ₀ -F ₆

表1通过眼睑曲线方程求取眼型指数范围Table 1 Calculate the range of eye shape index by eyelid curve equation

4眼型分类4 eye type classification

由于眼睛形状属于为人熟知的“一般”形状(也就是说，具体实例各不相同)，无法使用一种刚性形状模型来描述。所以把一个需要判别类型的眼型与不同类型眼型标准样本进行比较，如果需要判别类型的眼型与某一类眼型满足归类条件就认为这个眼型属于这类。Since eye shape is a well-known "general" shape (that is, it varies from instance to instance), it cannot be described using a rigid shape model. Therefore, compare an eye type that needs to be identified with different types of eye type standard samples, and if the eye type that needs to be identified and a certain type of eye type meet the classification conditions, the eye type is considered to belong to this category.

把需要判别类别的眼型分别与圆眼、标准眼、眯缝眼、细长眼这四类眼型标准样本通过公式(9)计算每类眼睛可能性C_i,作为归类条件之一。计算需要识别眼型的眼型指数E_r，通过公式(10)计算眼型指数差和D_i，作为归类条件之一。判断规则如下：Calculate the possibility C i of each type of eye by using the formula (9) to calculate the probability C _i of each eye type by using the eye types that need to be distinguished from the four types of standard samples of eye types: round eyes, standard eyes, squint eyes, and slender eyes, as one of the classification conditions. Calculate the eye shape index E _r that needs to identify the eye shape, and calculate the eye shape index difference and D _i through the formula (10), as one of the classification conditions. Judgment rules are as follows:

(1)当唯一i值同时满足C_i≥0.7，MIN(D_i)时，则待识别的眼型是第i类眼型。(1) When the unique i value satisfies C _i ≥ 0.7, MIN(D _i ) at the same time, the eye type to be identified is the i-th eye type.

(2)当多个i值同时满足C_i≥0.7，MIN(D_i)时，则MAX(C_i)的i值是第i类眼型。(2) When multiple i values simultaneously satisfy C _i ≥ 0.7, MIN(D _i ), then the i value of MAX(C _i ) is the i-th eye type.

(3)当i值无法满足C_i≥0.7，MIN(D_i)时，则无法识别这个眼型。(3) When the value of i cannot satisfy C _i ≥ 0.7, MIN(D _i ), the eye type cannot be identified.

${C C}_{i i} = = \frac{{W W}_{i i}}{{N N}_{i i}},, i i = = 11,, 22,, 33,, 44 - - - - - - ((99))$

其中W_i为一类眼型中上下眼睑相似度同时满足|ρ|≥0.95的个数，N_i为一类眼型中标准样本个数Among them, W _i is the number of the upper and lower eyelid similarities in a type of eye type that satisfy |ρ|≥0.95 at the same time, and N _i is the number of standard samples in a type of eye type

${D D.}_{i i} = = {Σ Σ}_{j j = = 11}^{N N} | | {E E.}_{r r} - - {E E.}_{i i j j} | |,, i i = = 11,, 22,, 33,, 44 - - - - - - ((1010))$

其中N为每类眼型中标准样本个数Where N is the number of standard samples in each eye type

附图说明Description of drawings

图1四种不同眼型采样点分布情况。(1为标准眼、2为眯缝眼，3为细长眼，4为圆眼)Figure 1 Distribution of sampling points for four different eye types. (1 is standard eye, 2 is squint eye, 3 is slender eye, 4 is round eye)

图2四次拟合效果图(a)。Figure 2 Quadruple fitting effect diagram (a).

图3四次拟合效果图(b)。Figure 3 Quadruple fitting effect diagram (b).

图4眼宽、眼高示意图。Figure 4 Schematic diagram of eye width and eye height.

图5方法流程图Figure 5 method flow chart

图6圆眼轮廓拟合效果图。Fig. 6 The effect diagram of round eye contour fitting.

图7标准眼轮廓拟合效果图。Figure 7 Standard eye contour fitting effect diagram.

图8眯缝眼轮廓拟合效果图。Fig. 8 The effect diagram of the contour fitting of the squint eye.

图9细长眼轮廓拟合效果图。Fig. 9 Fitting effect diagram of slender eye contour.

具体实施方式detailed description

本实验每类眼型选取无表情、眼部无遮挡正面人脸照各100张，方法流程如图4。In this experiment, 100 frontal face photos with no expression and eyes with no occlusion were selected for each eye type. The method flow chart is shown in Figure 4.

1使用最小二乘法拟合出眼睑轮廓曲线1 Use the least squares method to fit the eyelid contour curve

随机选取圆眼、标准眼、眯缝眼、细长眼这四类眼型标准图片各10张,使用圆眼轮廓曲线方程标准眼轮廓曲线方程细长眼轮廓曲线方程眯缝眼轮廓曲线方程分别拟合出圆眼、标准眼、细长眼、眯缝眼轮廓曲线，即眼型效果图如图5-8所示，从图中可以看出最小二乘法可以准确拟合出四种眼型的轮廓曲线。Randomly select 10 standard pictures of four types of eye types, namely round eyes, standard eyes, squint eyes, and slender eyes, and use the round eye contour curve equation Standard Eye Contour Curve Equation Slender Eye Contour Curve Equation Squint Eye Contour Curve Equation Fit the contour curves of round eyes, standard eyes, slender eyes, and squint eyes respectively, that is, the eye shape effect diagram is shown in Figure 5-8. It can be seen from the figure that the least square method can accurately fit four eye shapes contour curve.

2使用曲线切线斜率相似度判别眼型2 Using the slope similarity of the curve tangent line to judge the eye type

每次随机选取圆眼、标准眼、眯缝眼、细长眼这四类眼型标准图片各20张，进行三次实验，通过类内，类间图片两两进行计算归一化互相关系数，发现两个上、下眼睑|ρ|同时大于0.95时可以比较准确区分开不同类别眼型。如果两个眼型上下眼睑同时满足|ρ|>＝0.95说明这两个眼型属于同一类，否则相比较的两个眼型属于不同类型眼型。实验结果见表2，说明使用眼睑轮廓曲线切线斜率相似度判别不同眼型正确率约为74.3％。Randomly select 20 standard pictures of the four types of eye types, namely round eyes, standard eyes, squint eyes, and slender eyes each time, and conduct three experiments. The normalized cross-correlation coefficients are calculated by pairwise intra-class and inter-class pictures, and it is found that When the two upper and lower eyelids |ρ| are greater than 0.95 at the same time, different types of eye types can be distinguished more accurately. If the upper and lower eyelids of the two eye types satisfy |ρ|>=0.95 at the same time, it means that the two eye types belong to the same type, otherwise the two compared eye types belong to different types of eye types. The experimental results are shown in Table 2, which shows that the correct rate of distinguishing different eye types using the similarity of the tangent slope of the eyelid contour curve is about 74.3%.

表2曲线相似度眼型分类正确率Table 2 Curve similarity eye type classification accuracy rate

3使用曲线切线斜率相似度判别眼型3 Using the similarity of the slope of the curve tangent line to judge the eye shape

随机选取圆眼、标准眼、眯缝眼、细长眼这四类眼型标准图片各50张，通过计算眼型指数E_ij(i＝1，2，3，4，j＝1，2，…，50),眼型指数E_ij满足眼型指数范围见表1，实验结果见表3，说明使用眼型指数E_ij判别不同类型眼型准率为85.5％。但经过实验发现，不属于这四类眼型的眼型指数也满足这四类眼型指数范围，所以眼型指数不可以单独作为分类标准，只能在实验2的基础上对分类结果加强。Randomly select 50 standard pictures of four types of eyes, namely round eyes, standard eyes, squint eyes, and slender eyes, and calculate the eye shape index E _ij (i=1, 2, 3, 4, j=1, 2,... , 50), the eye shape index E _ij meets the eye shape index range shown in Table 1, and the experimental results are shown in Table 3, which shows that the accuracy rate of using the eye shape index E _ij to distinguish different types of eye shapes is 85.5%. However, after experiments, it was found that the eye shape index that does not belong to these four types of eye types also meets the range of these four types of eye shape indexes, so the eye shape index cannot be used as a classification standard alone, and the classification results can only be strengthened on the basis of Experiment 2.

表3眼型指数进行眼型分类正确率Table 3 Correct rate of eye type classification by eye type index

4使用曲线切线斜率相似度与眼型指数结合进行眼型分类4 Eye type classification using the combination of curve tangent slope similarity and eye type index

随机选取圆眼、标准眼、眯缝眼、细长眼这四类眼型标准图片各50张作为测试样本，随机选取圆眼、标准眼、眯缝眼、细长眼这四类眼型标准图片各20张作为标准样本。分类的正确率见表4，标准眼分类正确率最高为92％，圆眼与眯缝眼分类正确率最低86％，平均分类正确率为88.5％。Randomly select 50 standard pictures of four types of eye types, namely round eyes, standard eyes, squint eyes, and slender eyes, as test samples. 20 as a standard sample. The correct rate of classification is shown in Table 4. The highest correct rate of standard eyes is 92%, the lowest correct rate of round eyes and squinted eyes is 86%, and the average correct rate of classification is 88.5%.

表4曲线相似度与眼型指数结合分类正确率Table 4 The accuracy rate of the combination of curve similarity and eye shape index

Claims

1. An eye type classification method based on eyelid curve similarity and eye type index is characterized in that the eye type classification purpose is achieved by calculating eyelid contour curve similarity and eye type index difference between an eye type to be recognized and each type of standard eye type, and the eye type classification method comprises the following steps: :

let as standard samples, i ═ 1,2,3,4, four eye types, j ═ 1,2, …, N standard samples per eye type; unknow is the eye type to be identified; K-Unknow is the type of the eye type to be identified;

1) from Sample_ijIn extracting the direction of the feature pointsQuantity f_ij(x₀,y₀,…,x₁₂,y₁₂) Forming a feature point set F by all feature point vectors;

2) from each feature point vector f_ijCalculating to obtain a sampling point vector p_ij(x₀,y₀,…,x₄₇,y₄₇) Forming a sampling point set P by all sampling points;

3) vector p of each sampling point_ijDivided into upper and lower eyelids (x)₀，y₀，…，x_m，y_m)，(x₄₇，y₄₇，…，x_m+1，y_m+1) Two parts, if the inside and outside eye corner points are not included in (x)₀，y₀，…，x_m，y_m)，(x_m+1，y_m+1，…，x₄₇，y₄₇) In (x), the inner and outer canthus points are inserted_{Inner part}，y_{Inner part}，x₀，y₀，…，x_m，y_m，x_{Outer cover}，y_{Outer cover})，(x_{Inner part}，y_{Inner part}，x₄₇，y₄₇，…，x_m+1，y_m+1，x_{Outer cover}，y_{Outer cover})；

4) Use (x)_{Inner part},y_{Inner part},x₀,y₀,…,x_m,y_m,x_{Outer cover},y_{Outer cover})、(x_{Inner part},y_{Inner part},x₄₇,y₄₇,…,x_m+1,y_m+1,x_{Outer cover},y_{Outer cover}) Performing least square function fitting to obtain coefficient vector a of curve equation of upper eyelid and lower eyelid_ij(c₀,…,c₉)，c₀,…,c₄Is the coefficient of the upper eyelid curve equation, c₅,…,c₉Forming a coefficient set A for curve equation coefficients of the lower eyelid by curve equation coefficient vectors;

5) using vector p of sample points_ijAnd coefficient vector a_ijCalculating the slope of the tangent line at the sampling point of the curve to form a slope vector s_ij(n₀,…,n₄₇) Forming a slope set S by all slope vectors;

6) using vector p of sample points_ijAnd coefficient vector a_ijCalculating the eye index e_ijForming an eye type index set E by all eye type indexes;

7) computing the slope vector s of Unknow_unknow(n₀,…,n₄₇) Eye type index e_unknow；

8) Calculating s_unknowSimilarity C (C) with slope set S column vector₁,c₂,c₃,c₄) Calculate e_unknowSum of differences D (D) with E column vectors of eye exponent set₁,d₂,d₃,d₄)；

9) And distinguishing the class K-Unknow of Unknow by the similarity vector C and the difference sum vector D, and outputting the K-Unknow.

2. The eye type classification method based on eyelid curve similarity and eye type index as claimed in claim 1, wherein the vector p of sampling points in step 2)_ijIs calculated as follows:

the contour-based method uses the boundary information of the shape as the basis to determine the eye feature point F_i(i＝0…11)，(F₁₂Is the pupillary point, F₀Is the inner eye corner point, F₆Is the outer eye corner point, F₁—F₅Is the upper eyelid point, F₇—F₁₁Lower eyelid points) are sequentially connected to obtain an eye contour polygon; by pupil point F₁₂Is a central point, and the parallel X-axis line passing through the pupil point is 0⁰Line, sampling the equal central angle, equally dividing the center of the circle by 48, and intersecting the equally divided line with the polygon of the eye outline to obtain a sampling point P_i(i ═ 0, …, 47); through the inner eye corner point F₀To the outer eye corner point F₆Connecting line, connecting 48 sampling points P_i(i ═ 0, …, 47) divided into upper eyelid sampling points P_i(i ═ 0, …, m), lower eyelid sampling point P_i(i＝m+1，…，47)。

3. The eye type classification method based on eyelid curve similarity and eye type index as claimed in claim 1, wherein the upper and lower eyelid curve equation systems in step 4)Number vector a_ij(c₀,…,c₉) Is calculated as follows:

taking out radicalFitting the eye contour curve equation for 4 times into a formula (2), and solvingThe normal equation of (1);

respectively aligning the inner and outer canthus of round eye, standard eye, long and narrow eye and upper eyelid sampling point P_i(i ═ 0, …, m), lower eyelid sampling point P_i(i ═ m +1 … 47), substituting into the equation of law, the corresponding equation can be obtainedThe contour curve equation of the upper eyelid and the lower eyelid of the round eye is a formula (3), the contour curve equation of the upper eyelid and the lower eyelid of the standard eye is the formula (3), the contour curve equation of the upper eyelid and the lower eyelid of the elongated eye is the formula (3), and the contour curve equation of the upper eyelid and the lower eyelid of the slit eye is the formula (3);

[\begin{matrix} m & Σ_{i = 1}^{m} x_{i} & Σ_{i = 1}^{m} x_{i}^{2} & Σ_{i = 1}^{m} x_{i}^{3} & Σ_{i = 1}^{m} x_{i}^{4} \\ Σ_{i = 1}^{m} x_{i} & Σ_{i = 1}^{m} x_{i}^{2} & Σ_{i = 1}^{m} x_{i}^{3} & Σ_{i = 1}^{m} x_{i}^{4} & Σ_{i = 1}^{m} x_{i}^{5} \\ Σ_{i = 1}^{m} x_{i}^{2} & Σ_{i = 1}^{m} x_{i}^{3} & Σ_{i = 1}^{m} x_{i}^{4} & Σ_{i = 1}^{m} x_{i}^{5} & Σ_{i = 1}^{m} x_{i}^{6} \\ Σ_{i = 1}^{m} x_{i}^{3} & Σ_{i = 1}^{m} x_{i}^{4} & Σ_{i = 1}^{m} x_{i}^{5} & Σ_{i = 1}^{m} x_{i}^{6} & Σ_{i = 1}^{m} x_{i}^{7} \\ Σ_{i = 1}^{m} x_{i}^{4} & Σ_{i = 1}^{m} x_{i}^{5} & Σ_{i = 1}^{m} x_{i}^{6} & Σ_{i = 1}^{m} x_{i}^{7} & Σ_{i = 1}^{m} x_{i}^{8} \end{matrix}] [\begin{matrix} c_{0} \\ c_{1} \\ c_{2} \\ c_{3} \\ c_{4} \end{matrix}] = [\begin{matrix} Σ_{i = 1}^{m} y_{i} \\ Σ_{i = 1}^{m} x_{i}^{1} y_{i} \\ Σ_{i = 1}^{m} x_{i}^{2} y_{i} \\ Σ_{i = 1}^{m} x_{i}^{3} y_{i} \\ Σ_{i = 1}^{m} x_{i}^{4} y_{i} \end{matrix}] - - - (1)

wherein m is the number of sampling points

S^{*} (x) = Σ_{k = 0}^{4} c_{k}^{*} x^{k} - - - (2)

S_{t}^{*} (x) = Σ_{k = 0}^{4} c_{t k}^{*} x^{k}, t = 1, ..., 4 - - - (3)

Wherein t-1 represents round eyes, t-2 represents standard eyes, t-3 represents slender eyes, and t-4 represents squinting eyes;

c₀,…,c₄for upper eyelid curve equation coefficientsc₅,…,c₉For lower eyelid curve equation coefficientsc₀,…,c₉Form the coefficient vector a of the curve equation of the upper eyelid and the lower eyelid_ij(c₀,…,c₉)。

4. The eye type classification method based on eyelid curve similarity and eye type index as claimed in claim 1, wherein step 5) Sample_ijTangent slope vector s_ij(n₀,…,n₄₇) And step 7) slope vector s of Unknow_unknow(n₀,…,n₄₇) Is calculated as follows:

the eyelid tangent slopes of different types of eyes at the same position sampling point are different, while the eyelid tangent slopes of the same type of eyes at the same position sampling point are similar, so that the similarity between eyelid curves is described by using the slope characteristics of the eyelid sampling points; the geometric characteristics of the slope are obvious, the influence of zooming and displacement is avoided, and a good classification effect can be obtained; after derivation of the fitted curve equations of the upper eyelid and the lower eyelid, the value x of the sampling point is substituted into the formula (4) to obtain the value of the tangent slope of the curve of the sampling point of the upper eyelid and the lower eyelid;

S^{*^{'}} (x) = \frac{\partial S^{*}}{\partial x} = Σ_{k = 1}^{4} k \cdot a_{k} x^{k - 1} - - - (4) .

5. the eye type classification method based on eyelid curve similarity and eye type index as claimed in claim 1, wherein step 6) Sample_ijIndex of eye pattern e_ijAnd step 7) eye index e of Unknow_unknowIs calculated as follows:

the measurement of different types of eye forms by the measurement method shows that the eye form index E of different types of eye forms_iWith a clear distinction, the eye index E_iThe approximate range of the eye type indexes of different eye types is obtained by calculation according to the formula (5), see table 1, which proves that the eye type indexes can be used as eye type distinguishing characteristics,

E_{i} = \frac{| e_{h} |}{| e_{w} |}, i = 1, 2, 3, 4 - - - (5)

wherein,e_w＝F₀-F₆

6. the method of claim 1, wherein the similarity of eyelid curves and eye type is based onThe method for eye type classification, wherein the similarity C (C) in step 8₁,c₂,c₃,c₄) Is calculated as follows:

calculating the similarity of the curves of the two upper eyelids or the two lower eyelids through the formula (6), wherein the larger the rho is, the larger the similarity of the slopes of the tangents of the curves of the two eye types is, the higher the degree that the two eye types are the same type is, otherwise, the more different the two eye types are, the two eye types cannot be the same type,

ρ = \frac{Σ_{i = 1}^{N} X (i) Y (i)}{\sqrt{Σ_{i = 1}^{N} X {(i)}^{2} Σ_{i = 1}^{N} Y {(i)}^{2}}} - - - (6)

wherein Respectively two upper eyelid or two lower eyelid sampling points to be compared, N is the number of the sampling points,

calculating the similarity rho by simultaneously (6) the slope of the sampling point with the recognized eye pattern and the slope of the sampling point of the four-type eye pattern standard sample_ijC is calculated by the formula (7)_i，

c_{i} = Σ_{j = 0}^{N} ρ_{i j} - - - (7)

N is the number of sampling points, i is 1,2,3,4, four eye types.

7. The eye type classification method based on eyelid curve similarity and eye type index as claimed in claim 1, wherein the step 9) classification method of Unknow K-Unknow is as follows:

calculating the eye possibility C of each type of eyes by the formula (8) according to the eye type needing to be classified and four types of eye type standard samples, namely round eyes, standard eyes, squinting eyes and slender eyes_iAs one of the classification conditions, an eye shape index E for identifying an eye shape is calculated_rCalculating the eye-type index difference sum D by equation (9)_iAs one of the classification conditions, the judgment rule is as follows:

(1) when the unique i values satisfy C at the same time_i≥0.7，MIN(D_i) If so, the eye type to be identified is the ith eye type;

(2) when a plurality of i values satisfy C at the same time_i≥0.7，MIN(D_i) When it is, then MAX (C)_i) The value of i of (a) is the type i eye;

(3) when the value of i cannot satisfy C_i≥0.7，MIN(D_i) Then, the eye type cannot be identified;

C_{i} = \frac{W_{i}}{N_{i}}, i = 1, 2, 3, 4 - - - (8)

wherein W_iThe number N of the upper eyelid and the lower eyelid in the type of the eye simultaneously satisfies that | rho | is more than or equal to 0.95_iNumber of standard samples in one type of eye

D_{i} = Σ_{j = 1}^{N} | E_{r} - E_{i j} |, i = 1, 2, 3, 4 - - - (9)

Wherein N is the number of standard samples in each type of eye type.