Research On The Key Technologies For Low Quality Finger Venous Image Recognition

Human biometrics has always been a hot research topic in human beings.With the development of computer graphics and pattern recognition,the field is constantly evolving.Currently widely used biometrics face recognition technology and fingerprint recognition technology,but both belong to the biological characteristics of the body surface,the security is not enough protection.Vein identification technology,as the living body biological characteristics,just make up for the lack of body surface features,with higher security.Vein recognition technology to identify the most important finger vein technology.Finger vein image recognition technology in the practical process,due to the increasing number of users and the collection environment constraints,resulting in poor quality finger vein images collected from low-quality finger vein image extracted vein features robustness and significant Sexual decline is more obvious.Therefore,there is an urgent need for research on the key technologies for low-quality finger vein image recognition.In this paper,low-quality finger vein recognition in the key technologies from traditional methods and methods of convolutional neural network in two directions,the main innovations are:First of all,considering the influence of gray background on the gray gradient perceived by human eyes under different gray background conditions,the concept of human eye perceived gradient based on the human eye's critical visible deviation is defined.Based on this concept,the concept of human eye's perceived gradient adaptive adjustment Fractional Order Differential Enhancement Algorithm for Order.Because the traditional gradient does not consider the influence of the background brightness,this paper uses the relative gradient to replace the traditional gradient to measure the real change of the gray level.In the meantime,in order to improve the robustness of the relative gradient to noise,the traditional relative gradient is improved.The 8-neighborhood mean is used to replace the current pixel value,and its anti-noise effectiveness is proved.Finally,the proposed algorithm is used to enhance the finger venous images with uneven illumination and high contrast.The average gradient and the entropy and the visual enhancement are used to evaluate the performance of the proposed algorithm.Enhanced fuzzy enhancement algorithm and the traditional fractional differential enhancement algorithm with different orders are compared.The comparison results show the effectiveness of the proposed algorithm.Secondly,in order to solve the problem of low accuracy of vein boundary segmentation in the segmentation of low-quality finger vein images by the use of directional valley detection method,this paper deeply studies and discusses the directional valley detection and segmentation algorithm based on Niblack algorithm.Based on this,a directional valley detection finger vein image segmentation algorithm based on canny algorithm was proposed.First of all,the author analyzed the characteristics of local gradient of finger vein and obtained the conditions that veins boundary point should possess,and used it as the criterion of whether it is a real boundary point,combined with the non-maximal value obtained by canny algorithm The value suppression image performs boundary point correction on the result of the segmentation based on the direction valley detection.Finally,by comparing the segmentation results before and after the experiment,the segmentation result of the proposed algorithm is more excellent in the continuity of venous region and the veracity and accuracy of vein extraction,which shows that it is effective.Finally,the convolution neural network which is suitable for adaptive extraction of image features is studied.Aiming at the problem of low quality finger vein image recognition,a method of slider covering is proposed in image data set processing.This method uses a fixed-size zero-grayscale slider to slide-cover each original image.Each movement once sets the covered pixel value to 0,resulting in a new sample of the category to which the current original belongs.Because this method fully takes into account the fact that the whole image may exist in any location with low significant venous regions and shields these regions so that the trained network model can have a higher value for low-quality finger vein image recognition Robustness.Before and after using the proposed method,the testing recognition rate of convolutional neural network increased from 0.9590 to 1,indicating the effectiveness of the proposed method.