| Monitoring the dynamic changes of blood microcirculation is of great significance to the studies of biological functions and disease.Laser speckle contrast imaging technology can achieve two-dimensional speed imaging in a wide field of view,with the advantages of no scanning,fast imaging,and high resolution.However,in laser speckle contrast imaging,the movement of either the imaging device or the subject will cause significant interference to the estimation of blood flow speed,which hinders its practical applications.In this paper,the image registration algorithms to reduce the impact of motion on laser speckle blood flow imaging are investigated.The main research contents and results are listed as follows:(1)A new method of raw speckle image registration based on the LK optical flow pyramid algorithm is proposed to realize the effective registration of the non rigid motion,which significantly reduces estimation error of the blood flow speed and image blur caused by the movement of multiple raw speckle images.After registration,the peak signal-to-noise ratio(PSNR)between the standard deviation image of the laser speckle image and the standard deviation image of the reference image can be increased by 0.5 d B to 1.2 d B.(2)A new method of blood flow image registration is proposed by combining the LK pyramid algorithm and the demons algorithm,which has good correction effect on the whole large motion and local small motion of blood flow image.If the demons algorithm(50 iterations)is used alone for registration,the PSNR of floating image and reference image is increased about 4 d B,while the PSNR is increased about 6 d B by combining the LK optical flow pryamid and the demons algrithm(10 interations).(3)A registration software of laser speckle contrast imaging is developed,which uses GPU to accelerate thin plate spline transformation and demons algorithm.The calculation time of thin plate spline transform is reduced from 4.6 seconds to about 80 milliseconds.The calculation of demons algorithm is reduced from 3.2 seconds to about 450 milliseconds. |
