Superficial vein enhancement method based on multi-resolution residual fusion network

Abstract

The invention discloses a superficial vein enhancement method based on a multi-resolution residual fusion network. The superficial vein enhancement method comprises the following steps: simulating veins by randomly arranged black wires with different thicknesses; simulating different degrees of skin scattering through a frosted acrylic plate, and collecting images under different exposure time asa simulated vein data set; converting the simulated vein data set into a similar living body blood vessel imaging data set through style migration; randomly extracting a part from the quasi-living blood vessel imaging data set as a training set, and training by adopting a multi-resolution residual fusion network to obtain an enhancement model; and enhancing the real vein data by using the enhancement model to obtain a vein enhanced image. According to the invention, by training simulated vein data sets with different illumination conditions and different scattering degrees, an enhancement model with good robustness is obtained; and by fusing the characteristics of different levels in the network, information lost in the down-sampling process is compensated, and a better detail enhancementeffect is achieved compared with other networks.

Description

technical field [0001] The invention belongs to the technical field of biomedical imaging, in particular to a superficial vein enhancement method based on a multi-resolution residual fusion network. Background technique [0002] The acquisition and processing of images of superficial vein distribution is one of the important research topics in the field of biomedical imaging, and has important application value in the fields of medical diagnosis and treatment and biometric recognition. For example, in the field of clinical medicine, the feature extraction of superficial veins helps to improve the success rate of venipuncture; superficial veins, as living features, are universal and unique, difficult to forge or change, and are more user-friendly for biometric identification And safe. [0003] When collecting superficial vein images, near-infrared light scatters when irradiating the veins, uneven illumination, equipment defects, skin tissue texture, fat thickness, low temper...

AI Technical Summary

Problems solved by technology

Traditional unsupervised methods use filtering and histogram equalization, etc., such as the method used in "An Effective Quality Improvement Approach for Low Quality Finger Vein Image", the detected veins are not clear enough, the background is messy, and are greatly affected by imaging conditions; supervision The learning method needs a label that corresponds to the data one by one, but the superficial vein data lacks a database, and the label needs to be marked by professionals, which consumes a lot of manpower and financial resources, and the cost is high, so it is difficult to popularize and apply

Images