Refers
The ripple is the uneven texture on the front of the finger. Although fingerprint is only a small part of human skin, it contains a lot of information. Fingerprint features can be divided into two types: general features and local features. General Features
It refers to the features that can be directly observed by the human eye, including the basic pattern, pattern area, core point, triangle point, style line and number of lines. The basic pattern includes ring, bow, and spiral pattern. Local feature fingerprint
It refers to the final point, branch point, and turning point on the texture. These fingerprint feature points can be described using the following four features. (1) position: the position of a feature point is described by (x, y) coordinates. It can be absolute or relative to a triangle point. (2) direction: the direction of the local spine of the feature point. (3) classification: feature points are classified into the following types: endpoint, split point, split point, isolated point, cycle point, and short grain. The most typical endpoints and forks are 9-3. (4) spine: the spine (Di, AI) corresponding to the feature points ). The spine corresponding to the feature points are represented at the sampling points on the ridge line. The sampling points are represented by the distance di between the point and the corresponding feature point, the straight line between the point and the corresponding feature point, and the angle AI corresponding to the feature point direction. Fingerprint Recognition Technology generally uses the overall features of the fingerprint, such as the pattern and triangle, for classification, and then uses local features such as location and direction to identify the user's identity. Generally, we first find "Minutiae" from the acquired fingerprint image ), then, based on the features of the feature points, the digital representation of the user's living fingerprint-fingerprint feature data (one-way conversion, it can be converted from a fingerprint image to a feature data, but not from a feature data to a fingerprint image ). Because two different fingerprints do not produce the same feature data, pattern matching is performed on the feature data of the acquired fingerprint image and the fingerprint feature data stored in the database, calculate the similarity and obtain the matching results of the two fingerprints. Then, identify the user based on the matching results. In short, the fingerprint recognition technology first reads the fingerprint image, then uses the computer recognition software to extract the fingerprint feature data, and finally obtains the recognition result through the matching recognition algorithm. The basic principle is shown in Figure 9-4. According to the basic principles of fingerprint recognition, fingerprint recognition technology mainly goes through the following four steps: fingerprint image acquisition, fingerprint image preprocessing, fingerprint feature extraction and fingerprint feature matching. Each step of fingerprint recognition is described in detail below. (1) fingerprint image acquisition. Devices that obtain fingerprint images can be divided into three categories: optics, silicon crystal sensors, and others. The longest application of optical imaging equipment is based on the principle of total reflection of light. The Application of crystal sensors has recently appeared on the market. These flat containing micro Crystals use a variety of techniques to plot fingerprint images. A capacitive sensor is one of them. It is designed to capture fingerprint images through electronic measurements. The capacitive device can be combined with a 100 000 conductor metal array sensor. The outside is an insulating surface. When the user's fingers are placed on it, the skin forms the other side of the capacitive array. The capacitance value of the capacitor varies depending on the distance of the metal. Here, it refers to the distance between the ridge (near) and Valley (FAR. In addition to the above two categories, ultrasonic scanning is considered to be a very good technology in fingerprint imaging, but the price is too high and the volume is too large .. Generally, the acquired fingerprint images are stored in a gray scale of 256. (2) fingerprint image preprocessing. To obtain accurate fingerprint feature points, fingerprint image preprocessing usually involves image enhancement (Noise Removal by filtering), computational pattern, binarization, and refinement. The entire process is shown in 9-5. · Image enhancement. In general, the acquired fingerprint image has a lot of noise, such as finger staining, Scar, dry, wet, or torn, so how to obtain the fingerprint image, effectively filtering image noise is one of the challenges in fingerprint recognition technology. Through image enhancement, noise can be filtered out to enhance the contrast between ridge and valley. There are many image enhancement methods, but most of them are matched by filtering the image and the local direction of the ridge. The image is first divided into several small areas (Windows), and the local direction of the ridge is calculated on each area to determine the pattern, which can be processed by the spatial domain, or the local direction on each small window can be obtained through the frequency-domain processing after the rapid two-dimensional Fourier transformation. Then design a suitable, matched filter to apply all the pixels on the image (one of the space fields ). Based on the local direction of each ridge at each pixel, the filter should enhance the direction of the ridge in the same direction, and weaken any direction different from the ridge in the same position. Because the latter contains the noise that spans the ridge, the incorrect "bridge" perpendicular to the local direction of the ridge will be filtered out by the filter. · Calculate the pattern. The pattern describes the tangent direction of the spine or valley line of each pixel in the fingerprint image. It is a useful information that can be obtained directly from the source gray image, its computing has always been an indispensable step in fingerprint recognition technology. The pattern can also be seen as a way to change the representation of the source image of the original fingerprint, that is, the direction of a certain point on the line is used to represent the direction of the line. Generally, there are two kinds of pattern: one is the point pattern, indicating the direction of the spine of each pixel in the original fingerprint image, and the other is the block pattern, indicates the average direction of all elements in a point area of the original fingerprint image. The basic idea of calculating the pattern is to calculate a statistic (such as the gray difference and gradient) of each point in the original gray-scale fingerprint image ), determine the direction of the point (this block) based on the difference of these statistics in each direction. In practice, block pattern is often used, because the block pattern is often more noise-resistant than the point pattern, and the block pattern can reduce the calculation amount, which is conducive to modular processing. The block pattern can be obtained by the vertex pattern or by the Least Squares estimation algorithm. · Binarization. First
First, the adaptive automatic door limit can be obtained based on the assumption of the ridge and valley width of the fingerprint and the investigation of the local gray distribution. The Adaptive Threshold Value Selection Method is to first find the normal direction of the point, in the ideal
The normal average value can be used as the threshold value. However, considering the influence of noise, we should remove the average value of the vertex after the maximum and minimum values and add a modified value as the threshold value. The calculation formula is as follows: Where: R is the method to remove the maximum and minimum points of the average, T is the maximum and minimum points of the average, (T-R) 22 is the correction value, TT is the threshold. After the threshold value is selected, the point can be binarization and processed point by point. Second, in fingerprint images, considering that pixels in the same region should have a similar continuous gray scale, based on the gray-scale change, the adjacent gray-scale change is assumed to further confirm the foreground and background of the image element, the inconsistency of unclear fingerprints near the automatic door limit can be well ruled out. Third, in order to solve the limitations of binarization in image segmentation that the field of view is too small, and at the same time to process blurred areas and isolated noise, the generalized Laplace operator is used to filter the image.The experiment shows that this algorithm not only highlights the texture, but also retains the detailed features of the fingerprint, and greatly reduces the false information such as the broken seams and adhesion of the fingerprint. The binary operation changes a gray image to a binary image, and the image drops from the original 256 colors to 2 colors in the intensity hierarchy. After the image is binarization, subsequent processing will be easier. The difficulty of binarization is that, because not all fingerprint images have the same threshold, we generally cannot start with a simple intensity. In addition, the contrast of a single image changes. For example, because the hand is pressed in the center, therefore, a method called "Local adaptive threshold" is used to determine the threshold of local image strength. · Refinement In
The last process before fingerprint feature extraction is "refined ". Refinement is the process of reducing the width of the ridge to the width of a single pixel without affecting the topological connection of the source image. A good refinement method is to keep the original
The continuity of ridges reduces the impact caused by human factors. Human Factors mainly include glitch and short spine, which cause many pseudo features to be extracted. The advantage of the refinement method is to reduce memory space. It only needs to store the necessary structure information in the image. In this way, the data structure can be simplified during image processing. The key to refined definition is how to find the skeleton of the original image. The template matching method is usually used to process the Image Based on the image features of a certain pixel's local neighbor. Of course, there are also refined methods such as external Contour Calculation and neural networks. (3) extract fingerprint feature points. As shown in Table 9-1, Feature Extraction uses a 3*3 template to detect the location and type of feature points. m is the detected fingerprint feature point, N0 ,..., N7 is the adjacent point of feature m in the counterclockwise direction. If n8 = N0, M is the endpoint. If Where n8 = N0, M is the branch point, as shown in 9-3. Due to the influence of image noise and other factors, many pseudo feature points extracted from the above algorithm need to be deleted. The deletion of pseudo feature points can be divided into two steps: · If a section in a ridge chart is completely orthogonal to the direction of the local Ridge and its length is smaller than the value of T, the ridge will be eliminated. If a gap in the ridge is very short and no other ridge passes through, the missing ridge should be supplemented. · If the details in a small area form a cluster, only the one closest to the center is left. If the two segments are very close and there is no ridge in the middle, the two details are eliminated. After feature extraction, the following parameters should be retained for each feature: the X and Y coordinates of a feature point, the direction of a feature point, that is, the local ridge direction connected to the feature point, the type of the feature point, that is, the ending point, or the ridge connected to the feature point. The ridge connected to the feature point is represented by ridge spacing sampling along the ridge line. (4) Compare the feature points. In
During fingerprint input, even the same finger does not have the same fingerprint image, which may produce various deformation, such as translation and rotation. For effective matching, all kinds of deformation must be minimized, considering the fingerprint
All kinds of nonlinear deformation are usually radioactive, and fingerprint matching can be performed in the polar coordinate system. In addition, due to the existence of nonlinear deformation, it is difficult to find a feature point that is exactly the same as the location of the feature points in the fingerprint template. Therefore, matching
The algorithm should be elastic, that is, the error caused by Nonlinear Deformation in a certain range is allowed. Fingerprint feature matching uses the allow box to achieve elasticity. The allow box is a box around the feature point, as shown in 9-6.
|
