Real-time Processing And 3D Visualization Of OCT Fingertip Data

In today's society,biometric-based identity authentication is an important means to ensure information security and personal security.However,traditional epidermal fingerprints have a single modality and weak anti-counterfeiting capabilities.The dermal fingertips collected by Optical Coherence Tomography(OCT)contains subcutaneous tissues which have strong uniqueness and anti-counterfeiting ability.However,the measureed OCT data is complex and the data amount is very large,which is difficult to be processed in real time.This will also affect the subsequent real-time three-dimensional visualization and it is impossible to observe the fingertip subcutaneous immediately after the data acquisition is completed.In the process of OCT data processing,if CPU calculation is used,the CPU(Central Processing Unit)resources will be greatly consumed,and the processing speed of the system will be reduced.The emergence of GPU(Graphic Processing Unit)solves this problem.The famous graphics card company NVIDIA has designed a special GPU parallel computing toolkit for its mainstream graphics products,called CUDA(Compute Unified Device Architecture).CUDA can control hundreds of internal processors inside the graphics card as a thread processor to solve intensive data calculation problems.(1)This paper designs a GPU-based OCT data parallel processing method to accelerate the data processing process.In order to further analyze the data collected by OCT,this paper realizes a three-dimensional visualization of the processed OCT data,and clearly show the shape characteristics and spatial distribution of the subcutaneous tissues of the fingertip.Among various visulaization tools,the Ray-Casting algorithm is classic.But when the sampling step is small,its calculation amount is large,the drawing speed is slow,and it is difficult to draw a three-dimensional image in real time.In this paper,the Ray-Casting algorithm is improved on the sampling step size,which not only ensures the drawing quality but also improves the drawing frame rate.The main work and results of the thesis are summarized as follows: Using the graphics processor GTX 1050 Ti as the computing unit and CUDA as the programming tool,the parallelism of the OCT data processing algorithm is analyzed,and the parallel code of the OCT data processing is designed to realize the parallelization of OCT data processing,By testing the OCT data processing speed under CPU and GPU,the GPU is proved to have a significant acceleration effect on OCT data processing.(2)The processed data is saved into a three-dimensional volume data format for three-dimensional visualization.The visualization effects of the OCT fingertip data under different surface rendering and volume rendering algorithms are tested and compared.The highest-quality Ray-Casting is thereafter selected as the visualization algorithm in this thesis.(3)The Ray-Casting algorithm has a good rendering effect on the OCT fingertip data,but the sampling step is very small,and the calculation amount is very large,which is difficult to reconstruct in real time.To overcome the probelms,the Catmull-Rom curve is used to interpolate between the real sampling points to find the virtual sampling points,in order to improve the sampling rate,reduce the complexity of the Ray-Casting algorithm,and thus improve the rendering speed.By comparing the rendering frame rate of the fingertip subcutaneous tissues under different sampling steps,it is proved that the improved Ray-Casting algorithm can ensure the rendering quality while improve the rendering frame rate.Finally,by comparing the different three-dimensional visualization effects of true and false fingerprints,the broad application prospect of OCT in fingerprint anti-counterfeiting is proved.