Finger vein recognition method based on multi-scale HOG and SVM

Abstract

The invention relates to the technical field of finger vein recognition, and specifically relates to a finger vein recognition method based on a multi-scale HOG and a SVM. The method includes following steps: acquiring multiple original finger vein images of many people and dividing the images into a training set and a test set; respectively performing extraction of region of interest, preprocessing, and image enhancement processing on all the images in the training set and the test set to obtain finger vein line information, and extracting multi-scale HOG characteristics of the line information; then transmitting the multi-scale HOG characteristics extracted from the images of the training set to an SVM multi-class classifier for training to obtain a trained SVM multi-class classifier; and finally transmitting the multi-scale HOG characteristics extracted from the images of the test set to the trained SVM multi-class classifier for recognition to obtain a matching result. According tothe finger vein recognition method based on the multi-scale HOG and the SVM, the finer vein recognition method based on lines and the finger vein recognition method based on non-lines are combined sothat accurate characteristic expression of finger veins can be realized, the recognition precision is high, and the robustness is high.

Description

technical field [0001] The present invention relates to the technical field of finger vein recognition, more specifically, to a finger vein recognition method based on multi-scale HOG and SVM. Background technique [0002] Existing finger vein recognition methods can be divided into two categories, namely, non-textured methods and texture-based methods. The non-textured method extracts features in the entire ROI of the finger vein image, without distinguishing between textured and non-textured regions. The texture-based method first extracts the vein texture in the region of interest, and then uses the extracted texture information to measure the similarity. Commonly used non-textured methods include LBP, minutiae, principal component analysis, and superpixel features. These features use the information of non-textured regions during matching, which can easily increase the similarity of heterogeneous images, resulting in false acceptance and poor recognition performance. D...

AI Technical Summary

Problems solved by technology

Commonly used non-textured methods include LBP, minutiae, principal component analysis, and superpixel features. These features use the information of non-textured regions during matching, which can easily increase the similarity of heterogeneous images, resulting in false acceptance and poor recognition performance. Difference

Although texture-based methods such as linear tracking, maximum curvature point, Gabor filter, and average curvature only use the information of the vein area, their recognition performance is still not ideal. The main reason is that the extracted texture is incomplete and noisy, while matching method is not robust

Images