Superficial Vein Image Enhancement Technology Research

Because the superficial veins 5～10mm below the skin cannot be observed directly by naked eyes,it is difficult to perform venipuncture by traditional visual method.The one-time success rate is low,especially for infants,women,elderly patients,obese people,etc.To realize vein visualization by near infrared imaging can effectively assist medical staff to perform accurate venipuncture,and even be suitable for disease diagnosis and vein recognition.However,due to the shortcomings of near-infrared vein images,such as low contrast,uneven gray level,a great deal of noise and unclear vein structure,image enhancement technology is of great significance for the realization of vein visualization and related applications.This paper studies two methods of superficial vein image enhancement,and the main contents are as follows:(1)The improved algorithm of the vein image enhancement method based on the maximum curvature method(MC)and Hessian matrix: on the one hand,CUDA(Compute Unified Device Architecture)acceleration and structure simplification are carried out for the image enhancement method combined with MC and Hessian.Before the four different vein centerline images are separately enhanced by Hessian,they are merged to form an overall image,and then the Hessian is performed only once.The operation speed is increased to 7.8 times,and the enhancement effect is improved to a certain extent.On the other hand,the application of a multi-scale Hessian matrix,which fuses the response output of vein vessels of different sizes,can effectively improve the image contrast,enhance thin veins,obtain more vein details and the more complete structure.(2)Superficial vein enhancement method based on multi-resolution residual fusion network: for the lack of high-quality reference images of vein images,the simulated vein dataset is collected,and an enhancement model is trained as the pre-trained model.The dataset more similar to real vein images is generated by style transfer and continues to be trained for the enhancement model to enhance real vein images.This paper proposes a multi-resolution residual fusion network structure,which can effectively use the characteristics of different layers to make up for the information lost during the downsampling process.This method can enhance superficial veins of different individuals under different illumination,and can effectively enhance deep,marginal and small veins.