A NEW FACE FEATURE POINT MATRIX BASED ON GEOMETRIC FEATURES AND ILLUMINATION MODELS FOR FACIAL ATTRACTION ANALYSIS

. In this paper, we propose a 81-point face feature points template that used for face attraction analysis. This template is proposed that based on the AAM modelaccording to the geometric characteristics and the illumination model. The experimental results demonstrate that, the attraction of human face can be analyzed by the feature vector analysis of human face image quan-tiﬁcation and the inﬂuence of light intensity on the attraction of human face. By taking the appropriate algorithm, the concept of facial beauty attractiveness can be learned by machine with numeric expressions.


1.
Introduction. Beauty is a universal part of human experience, and the perception of facial beauty or attractiveness is one of the most common human activities in our daily life. Facial beauty delights our sight and provokes pleasure in our mind, but, what is beauty? Is there any "beauty code" existing objectively?Mounir Bashour (2006) argues that human attraction is a visual enjoyment of face-to-face encounters, and its operation is defined as a static two-dimensional face visual feature that gives viewers a visual enjoyment [11].Rubenstein,Langlois and Roggman created averaged faces by morphing multiple images together and proposed that averageness is the answer for facial attractiveness [6,12].
At present, to study face attractive is mainly from the field of psychology and medical at domestic and foreign. To study the beautiful face attractive has just risen in recent years by using computer information processing technology.The ultimate goal of machine intelligence is to let the computer have the same intelligence as people, including perception, reasoning, judgment, identification and other aspects of the ability [9].Since people have a cognitive ability for beautiful appeal to the face, is it also possible to let the computer through the machine to learn the means to get the same intelligence? We think this is a basic question of whether the concept of "beautiful attraction about human face" can be learned.
In [14], based on the AAM model, an algorithm for fast and accurate fitting of AAMs under field conditions is proposed.In [1], the new method of distinguishing response mapping (DRMF) is proposed to improve the shortcoming of CLM framework. In [3], proposing several geometric features that characterize the attraction of Chinese women's faces;On the basis of this, the paper analyzes the attractiveness of oriental people from the geometric features, and then analyzes the influence of the illumination model on the attraction of the human face, and provides a reference for the judgment of the human attraction.In the past, the data were mostly used by Westerners, and there were few reports on the beauty of the face of the Orient. In this paper, the Orient people face image data, focusing on the Oriental face to analyze the attraction.
2. AAM. The current research results show that the model-based location method has better positioning effect and robustness.Model-based feature point positioning method The two most typical models currently used in the most widely used are the active shape model (AAM) and the active visual model (ASM) [10].
AMM are generative, parametric models of a certain visual phenomenon that show both shape and appearance variations [4]. First, Learning the shape model requires consistently annotating a set of landmarks [ x 1 , y 1 , x 2 , y 2 , ... x u , y u ] across training images , These points are said to define the shape of each object. Next, Procrustes Analysis is applied to normalize and remove similar transformations from the original image [5]. Finally, PCA is applied on these shapes to obtain a shape model defined by the mean shape and shape eigenvectors.
Assume that we are given a new similarity-free shape s = (x 1 , y 1 , x 2 , y 2 , ... x u , y u ) T and shape variation is expressed by a linear combination of a mean shape s 0 and n shape basis vectors s i , Then, the model can be used to represent s as where s 0 is the mean shape, p i denotes the i th shape parameter, and p = {p 1 , p 2 , x 2 , ... p m } is the shape parameter vector for the input face image. The principal component analysis of all the textures is obtained where p is the average texture, p g is the transformation matrix formed by the texture components of the texture components calculated by PCA, and p g is the statistical texture parameter that controls the texture change. The appearance variation is expressed by a linear combination of a mean appearance A 0 (x) and n appearance basis vectors A i (x) as Where λ denotes the i th appearance parameter, and λ = {λ 1 , λ 2 , ...λ n } is the appearance parameter vector for the input face image. The pixel value of the input image at pixel W (x; p) is I(W (x; p)). The sum of squares of the difference between A(x) and I(W (x; p)) is minimized, that is, the error function: where q is the global pose parameter vector including the scale,rotation, and horizontal/vertical translation. The following figure shows using the AAM algorithm detected the characteristic points by the Westerners and the Orientians.
3. The algorithm proposed in this paper.
3.1. Geometric model. Geometric features are a significant problem in quantitative description of human attraction. It has the same impact on the West and the East for the distance of facial features and the proportion of the beautiful face. The geometric methods used by predecessors are less likely to take into account the direct use of facial features as a feature to predict the face of beauty, we believe that facial features also indirectly reflect the face of rich geometric information, such as eye size, facial features , Forehead width and so on.  The distance on both sides of the forehead Table 1. 7-dimensional distance feature vector description.

3.2.
Light illumination model. When the light emitted from the light source is irradiated onto the surface of the object, the reflected light and the transmitted light can stimulate the human eye to produce a visual effect. The intensity of reflected light and transmitted light determines the degree of shading on the surface of the object [13]. From the visual effect, in the illumination model, assuming that the light emitted by the light source is white and the object is opaque, the color of the surface of the object is reflected only by the light Decision. In many cases, the content of the image does not change, but the change of light has changed the effect of the image. The illumination image of different regions in the face image is different. The size of the light intensity affects the judgment of the attraction of the face.
According to Lambert's law, the intensity of diffuse reflected light reflected on the surface of an ideal diffuse object is proportional to the cosine of the angle between the incident light and the surface normal of the object [2].
Where I is the brightness of the diffuse reflected light at the illuminated point P , I l is the incident light intensity emitted by the point light is the incident light intensity emitted by the point light source, k d (0 ≤ k d ≤ 1) is the diffuse reflectance of the surface of the scene, and θ is the incident The angle between the light and the surface normal vector. (5) can be expressed as the following vector form if the unit vector of the surface of the scene is N at the point of irradiation P and the unit vector of P to the point source is L, In the local light illumination model, we often assume that the ambient reflected light is a uniformly diffuse light and uses a constant to represent its intensity. In this way, the Lambert diffuse light illumination model can be written as: Where I a is the incident flood light intensity and k a is the diffuse reflectance of the surface of the object to flood light. It is worth noting that equation (7) does not reflect the distance attenuation effect of light. It is well known that the propagation of light is attenuated from the square of the square, that is, the intensity of the incident light somewhere is inversely proportional to the square of the distance between the point and the light source. we can use the following Lambert diffuse reflection model to simulate the various attenuation effects of light: 4. Experiments and results. In this paper, it is quantifiable the abstract concept of human attraction. Through the geometric features and the illumination model, the human face can be quantitatively predicted by machine learning. A total of 89 samples of the experiment, this article only selected 6 as a description. According to the statistical evaluation of human attraction, it is more attractive when the height of the nose is approximately the same as that of the ear. The closer the face triangle is to the nearest 60 degrees, it is more attractive. Figure 3, the new forehead feature points 3, the tip of the nose, the amount of heart, and both sides of the forehead to take the characteristics of a triangular area, the triangular area closer to the triangle, especially the tip of the nose and the angle of the eye closer to 60 degrees, The face is more attractive. Table 2 can be seen in the nose and ears in the face of the attraction, when the slope of −0.1 < k < 0.1, more attractive, k > 0.1, slightly less attractive. In the experimental sample, Figure 2 and Figure 4 are more attractive.   Table 2. The slope of the nose of the ears.