Research On Palmprint Image Preprocessing Algorithm For Real Scene

With the development of science and technology and the progress of society,traditional authentication such as passwords cannot meet the requirements of convenience and reliability to a large extent.Palmprint recognition is a new biometric technology in the past decade.Because palmprint images have the advantages of stable characteristics,simple collection methods,and easy acceptance by the public,palmprint recognition has broad research prospects.The palmprint recognition system includes three major steps: image acquisition,image preprocessing,feature extraction and matching.According to the different ways of image acquisition,it can be divided into two categories: contact palmprint recognition and non-contact palmprint recognition.Current palmprint recognition research is mostly a non-contact method.The user does not need to touch the plane of the sensor to capture the hand image,which greatly improves the convenience of use.At present,the research of non-contact palmprint recognition technology mainly focuses on the two aspects of palmprint image preprocessing and palmprint feature extraction and matching.Although today's palmprint recognition system has achieved high recognition accuracy,most of the collected images are palmprint images under a single background,which is very different from the images collected in real scenes.Using palmprint image preprocessing algorithms under a single background is difficult to meet the requirements in real scenes,which limits the promotion and application of palmprint recognition.Therefore,studying palmprint image preprocessing algorithms in real scenarios has great research value and reality.The research of palmprint image preprocessing algorithm mainly focuses on palm foreground segmentation and palmprint ROI(Region of Interest,ROI)positioning.This thesis studies the palmprint image preprocessing algorithm under realistic complex background.The main work is as follows:1.In order to get a better foreground segmentation effect,this thesis proposes to use the target detection algorithm to detect the palms in the image before foreground segmentation.And in order to facilitate the deployment of the algorithm on mobile or embedded devices,a palm detection model based on a lightweight convolutional neural network is proposed and experiments are conducted on a self-built palm data set.Experimental results show that compared with other lightweight network models,the proposed lightweight palm detection model has a smaller amount of parameters and calculations while maintaining considerable accuracy.2.In order to solve the problem of color deviation caused by light interference in images collected in real scenes,a palm foreground segmentation algorithm based on Gaussian model and white balance is proposed and experimental analysis is carried out on a self-built palm data set.Experimental results show that compared with other classic foreground segmentation algorithms,the proposed palm segmentation algorithm has the highest segmentation success rate.3.Aiming at the problem that the position and direction of the palm are not fixed in the images collected in real scenes,a palmprint ROI positioning algorithm based on direction correction is proposed.Compare and analyze the positioning stability and positioning success rate with other two mainstream ROI positioning algorithms.Experimental results show that the proposed ROI positioning algorithm has the best positioning stability and the highest positioning success rate compared with the other two mainstream ROI positioning algorithms.