The technology of fingerprint feature extraction from fingerprint image in fingerprint time attendance access control machine

The fingerprint features commonly used by fingerprint attendance Access Control machines in fingerprint identification are nodes, singular points and lines, etc. The nodes mainly include endpoints and points, and singular points include core points and triangle points. The extracted fingerprint features are used for fingerprint matching. The key technologies involved in fingerprint feature extraction mainly include texture direction calculation, texture frequency calculation, core point and triangle point detection, fingerprint segmentation, fingerprint enhancement, texture extraction and refinement, node extraction and filtering, and texture count. calculation etc.

The calculation of the texture direction is the basis of fingerprint identification. Most of the algorithms in fingerprint identification are based on the direction, such as frequency calculation, texture tracking, detection of core points and triangle points, fingerprint segmentation, fingerprint enhancement, node alignment, etc. Most of the algorithms are based on orientation. The texture direction calculation method is based on the gray level between pixels, compares each 2x2 block with four edge templates to extract the direction of the pixel block, and then makes an average estimate based on a larger area to calculate the more If it is difficult to determine the direction, the planer uses the grayscale alignment method to calculate the texture direction, discretizes the texture direction into 16 directions, and calculates the grayscale consistency of each pixel in each direction. Take the direction with the best consistency as the direction of the home, and calculate the grayscale change along each direction, the grayscale change along the grain direction is the smallest, and the grayscale change along the direction perpendicular to the grain is the largest. Convert into textured pixels and non-textured pixels, discretize the texture direction into 16 directions, and calculate the consistency of the pixel type of each pixel in each direction, use the projection method to calculate the texture direction, and divide the fingerprint image into a size of 32n32 block, and calculate the projection of each block in different directions, take the direction with the largest projection variance as the vertical direction of the texture, and use a hierarchical neural network to calculate the direction field. Currently, the most widely used texture direction calculation method is based on gradient. The method is poor. This method calculates the gradient vector of the fingerprint image at each pixel. The direction of the gradient vector represents the fastest grayscale change of the fingerprint image along this direction at the pixel, and the size of the gradient vector represents the speed of the grayscale change. The pixel gradient at the edge of the texture in the image is larger, the texture direction calculated by this method is basically determined by those pixels with a larger gradient, and the gradient direction of the image at the edge of the texture is basically perpendicular to the texture direction. The texture direction of each area is based on all the cities in that area.
The gradient vector of the pixel is calculated, and the gradient vector direction of the edge image cable on the left and right sides of the texture is just opposite, in order to avoid mutual cancellation in the calculation. In the calculation, the antimony degree vector is squared, and the squared gradient vector of the edge pixels on the left and right sides of the texture will point in roughly the same direction, and then the average direction of the squared gradient vector is calculated. Because the direction of the squared gradient vector is twice that of the squared gradient vector, 1 ni of the almost average direction of the nearly squared gradient vector is the vertical direction of the texture. The problem with the gradient-based method is that when the When the gradient direction of most edge pixels is not suitable for the direction of the texture, the wrong direction will be calculated as the original image, which is the direction estimation result. There are many wrong direction estimations in the elliptical area, which is the gradual enlargement of the local area image, from When zooming in on the national image, it can be seen that the gradient vectors of most of the edge pixels are not perpendicular to the direction of the texture, thus resulting in a wrong direction estimation. The method can correct the wrong direction of relative isolated work, but when the wrong direction in a certain area is the majority, not only the sense of wrong direction can be corrected. The correct direction will be incorrectly corrected, and the direction near the core point will often deviate from the true direction after low-pass filtering. When the curvature of the texture near the core point is large.