Research On De-noising Algorithm Of Vein Image Under Finger And Mirror Contamination

Finger vein recognition system is an efficient and safe identity authentication system.However,in the process of image acquisition,dirt except finger vein will absorb near-infrared light,which will form redundant interference information in finger vein image.Such information can be defined as pollution noise in actual use.In special application scenarios,due to the user factors(finger with molting,finger with skin cracks),environmental factors(stains attached to the mirror,dust easily attached to the mirror),will make the collected finger vein image carry noise.Due to the existence of these noises,the quality of finger vein image is reduced,and the feature extraction of finger vein image is affected.Finally,the matching performance of finger vein recognition technology in special application scenarios is reduced.Firstly,the existing image noise detection algorithms can not accurately extract the noise distribution and the existing image filtering denoising algorithms do not make good use of the characteristics of noise distribution,resulting in poor noise removal effect.A finger vein image denoising algorithm based on sparse structure noise detection is proposed.Secondly,in order to solve the problem that the widely used Conditional Generative Adverse Nets denoising model does not have accurate texture constraints,which leads to the problem that texture features of images are easily destroyed in the process of processing,a finger vein denoising algorithm based on Custom Sample-Texture Conditional Generative Adverse Nets is proposed.The proposed algorithm uses accurate texture constraints to guide the process of image denoising,which can protect the vein texture information more effectively while removing the noise.It has a good reference value for the noise detection and noise removal of vein and blood vessel images.The specific research contents are as follows:1.A denoising algorithm based on sparse structure noise detection is proposed.Firstly,the RPCA model is established based on the sparse structure of finger vein.By solving the model,we can get the foreground image with sparse targets.The next step is to extract noise distribution image by threshold segmentation.Through the above methods,we solve the problem of inaccurate detection of noise blocks with non-uniform distribution based on neighborhood correlation noise detection.At the same time,repair priority rule and adaptive selective filter template are established based on the extraction results to remove noise In addition,it solves the problem that the traditional spatial domain de-noising algorithm does not make full use of the characteristics of noise distribution.The experimental results show that the rejection rate of this method is 11.10%,5.95%and 3.64%lower than that of non processing,anlm denoising algorithm processing and FG noise detection denoising method processing,improving recognition performance of finger vein image with noise.The theory and analysis show that compared with the denoising algorithm based on ANLM and FG,the finger vein image denoising algorithm based on sparse structure noise detection can extract more complete noise distribution image,and the vein area covered by noise is more coherent after repair,and the texture information restoration result is better.2.We propose a finger vein de-noising algorithm based on Custom Sample-Texture Conditional Generative Adversarial Nets(CS-TCGAN).The proposed algorithm effectively protects the texture features while removing noise.Firstly,the proposed algorithm uses texture loss,adversarial loss and content loss as constraints,which leads to a better de-noising performance on finger vein image with blurred texture.Secondly,the dimension preserving structure is adopted in the generator network to minimize the problem of details lost caused by de-convolution.Lastly,the noise distribution of finger vein images obtained in the practical application has been investigated to generate the training dataset.Specifically,the training dataset has been established by combining Poisson noise,salt/pepper noise,Gaussian noise and speckle noise.The experimental results illustrate that the performance of the proposed algorithm is better than the traditional filtering de-noising approach and the widely used CGAN de-noising algorithm.