Biometrics is currently the most convenient and safe identification system. Biometrics is a solution that relies on the physical characteristics of the human body for identity verification. Compared with traditional biometric technology, face recognition is generally favored by people due to its many advantages such as easier, more accurate, economical and good scalability, and is widely used in various aspects such as security verification, monitoring, and control. . But so far, an ideal system that can meet people's needs has not yet appeared. According to the FERET'97 test report, the current face recognition algorithm is very poor in adapting to different cameras, different lighting conditions and age changes. The FRVT'2000 evaluation results show that the performance of the face recognition system has improved compared with the FERET'97 test, but its recognition performance is still far from people's attention to various conditions, such as light, aging, distance, and posture. Expectations are far away. This article only considers the frontal static grayscale image of a single person, focusing on the detection and positioning of the face image, and does not consider the head tilt, rotation, wearing, and shielding, and the head tilt does not exceed 15°. The research work of this paper is mainly to propose a method of combining face templates and facial features for face detection, and to improve the existing face detection and positioning methods, and then extract facial features, and implement it on this basis A demonstration software for an intelligent identification system was created. The method used in this paper is mainly based on a parameterized elliptical face template and a face positioning method based on the combination of the geometric characteristics of the eyes and mouth, and the geometric characteristics of the eyes, nose, and mouth. The face database is a method of face recognition under supervision of classification and statistics. The face database used is BioID face database. The automatic face recognition system includes the following main technical links. The first is face detection and localization, that is, finding the face and the location where the face exists from the input image, and segmenting the face from the background, and then aligning Characteristic extraction and recognition of face images. The research of these two links is very independent. In many specific situations, the work of face detection and localization is relatively simple, so the "feature extraction and recognition" link has been more extensively and in-depth studied; and in recent years, as people are more and more concerned about various complex situations Face detection and localization have only received more attention under the automatic face recognition system. The standard for evaluating an automatic face recognition system is the false recognition rate, which means that someone is mistakenly recognized as another person, and the other is the false alarm rate, which means that someone is recognized as this person. There is a contradiction between the two, so some compromises are often needed in practical problems. This is also applicable to the feature extraction and recognition process, but for face detection and positioning, we generally require the misrecognition rate to be as low as possible, because this can ensure that the person to be recognized will not be lost at this step. The commonly used face recognition experiment library is the FERET library of the US military as the most authoritative. In addition, libraries such as MIT and ORL can also be used to verify the capabilities of the algorithm in certain aspects. Currently, there is no image library specifically for testing face detection and positioning algorithms. 4.1 Detection of facial features Feature detection is a key link in the face recognition intelligent system, because features such as eyes, nose, and mouth concentrate part of the information of the face. For face modeling, it is not only necessary to detect these features, but also to accurately locate them. Assuming that the posture of the face is relatively positive, the features of the face are the areas where the horizontal edges are concentrated. We can extract the horizontal edges at low resolution, and then find the connected areas rich in these edges in the face area as the candidate area of ​​the face feature. The location and size of these regions are not accurate, because the edge detection itself is prone to position offset, and the size of the connected region also changes with the threshold. Therefore, the above results need to be further revised. For a frontal face, there is a priori constraint relationship between features such as eyes and mouth and the size of the entire face. This is the constancy of the face structure, so we use these constraints to determine the size of the feature area. 4.2 Positioning of the eyes and mouth You can make full use of prior knowledge, express the eyes and mouth as piecewise polynomial curves, and then use the deformed template to get the accurate contour. The eye contour model is shown in Figure 1. It consists of four quadratic curves. There are 8 parameters of the curve: the center of the upper eye (x0, y0), the inclination angle θ of the straight line connecting the inner and outer corners of the eye, the height of the upper and lower eye h1 and h2, the distance wl from the inner corner of the eye to the center of the upper eye, w2 from the outer corner of the eye to the center of the upper eye, and w3 from the inner corner of the eye to the center of the lower eye. Figure 1 Eye contour model The choice of evaluation function is the key. The eye itself does not have consistent color information, but the edge information is richer. Therefore, by first extracting the edges of the image, and then performing the closing operation of the appropriate scale, a single high-brightness area can be formed inside the eyes. In the processed image, the inside of the eye is a high-brightness area, and the outside is a low-brightness area. Therefore, the evaluation function can be defined as shown in equation (1). Where D represents the eye area, represents the band-shaped area outside the eye contour, and represents the band-shaped area inside the contour. The eye area obtained according to the 4-segment curve expression can initialize the contour of the eye, and then use the steepest descent method to iterate until the contour converges. The model of the mouth contour is shown in Figure 2. Composed of two quartic curves, the curve has 7 parameters: the center of the mouth (x0, y0), the inclination angle θ of the straight line connecting the corners of the mouth, the heights h1 and h2 of the upper and lower half of the mouth, the distance w from the corner of the mouth to the center of the mouth, and The quartic coefficients q1 and q2 of the upper and lower half mouths. Figure 2 The model of the mouth contour The evaluation function of the mouth is relatively easy to determine, and the mouth can be distinguished by the segmentation of lip color and skin color. The expression of the evaluation function is shown in equation (2). The meaning of each symbol is similar to the eye model. 5.1 System workflow The workflow of this system is as follows: After the picture is input, the image preprocessing is performed first, and then the ellipse template is used to initially locate the face. After correcting the direction of the face, look for the eyes and mouth based on the shape features, and verify the face area based on the correlation, and finally output the detection and positioning results. In the face recognition stage, feature extraction is performed on the image output from the detection and positioning, and then feature value statistics are performed under manual supervision. Then the criterion is constructed according to the template parameters and characteristic statistical values. For the input image to be recognized, the template parameters and feature values ​​are detected, located and calculated, and then the criteria are discriminated and classified and the recognition results are output. 5.2 Face detection and localization results After boundary detection, the result of face detection and positioning is shown in Figure 3. Figure 3 Face localization results when face shape, eyes and mouth features are considered at the same time 5.3 Detection and identification results This article uses 100 static gray images (BioID_0000.pgm-BioID_0099.pgm) in the BioID face database as materials to conduct experiments. The experimental results are shown in Table 1. Table 1 Detection and identification results 5.4 Result analysis Among the above three methods, the positioning method combining the features of the eyes, eyebrows and mouth with the parametric ellipse template shows better performance. When the method of image segmentation based on skin color is used to locate the face, the skin color is affected by the ambient lighting and the presence of objects similar to skin color in the background, resulting in poor positioning results. When the threshold value is changed, it has little effect on the judgment result, which shows that: first, the mixed skin color and background color, and the influence of strong light are the main factors that cause misjudgment; second, in one of the above problems In an "ideal" state that is too serious, the skin tone model can indeed function effectively to offset the effects of appearance and posture. The improvement of the skin color model should focus on the processing of the illumination and the separation from the background. The use of templates for detection and positioning is an effective method to separate background interference. When the elliptical template is used for detection and positioning, the detection accuracy has been improved to a certain extent, but in general it is still difficult to be satisfactory. When testing its performance by changing the parameters of the ellipse, its adaptability to parameter changes is high. After analysis, it can be considered that the complex background directly affects the effectiveness of the ellipse template. The poor adaptability to parameter changes is mainly due to the fact that faces in the image database are mostly oblong, with little difference in parameters. Therefore, the effect of detecting and locating the face only through the elliptical template is not satisfactory. On the basis of the parametric ellipse template, the parameter description of the features of eyes, eyebrows and mouth was added, and then the improved template was used for detection and positioning, and good results were obtained. Moreover, it also has a better performance when distinguishing target characters by adjusting parameters. After analysis, the improvement of the detection and positioning success rate is due to the effective description of the shape parameters of the facial organs, which filter most of the interference in the background and achieve better results. When adjusting the shape parameters of the facial organs, the model achieved a better distinguishing effect, which can be considered as a true response to the personal characteristics of each target person. The problem that still exists is that when the face is tilted, or there are glasses, large pieces of hair covered, etc., it will have a serious impact on the detection and positioning effect. The trend of face detection and recognition research is to use a variety of clues (hair, skin color, organ, contour, template, etc.), integrate multiple classification methods (mixed Gaussian model, probability model, neural network and support vector machine, etc.), heuristics Information and statistical learning methods are combined. In short, due to the non-rigidity of face objects, as well as the influence of various changing factors such as posture, illumination, occlusion, and real-time requirements, high-performance face detection is still a difficult problem.
Aluminum Laptop Cooling Stand material is fast and durable, and has a delicate texture. Laptop Stand Aluminum Adjustable is lightweight, portable and easy to store, stable without shaking, the bottom is hollowed out for heat dissipation. Laptop Stand Black say no to the cumbersome, anodized sandblasted, aluminum alloy material is formed in one piece. Ergonomic Aluminum Laptop Stand all aluminum alloy material, mini foldable, silicone non-slip, six-level height adjustment, hollowed out heat dissipation, compatible with multiple sizes. Laptop Stand And Risers is creative bracket, dedicated to create fashionable metal, the appearance is beautiful and generous, exquisite and beautiful, which is unforgettable.
Phone Holder Stand,Magsafe Stand For Iphone 12,Iphone Magsafte Stand,Phone Holder Stand Mobile Shenzhen ChengRong Technology Co.,Ltd. , https://www.laptopstandsupplier.com
(1)
(2)
1 Research background