The Study Of Human Face Image Registration Method Applying To Robot Assisted Surgery

The technology of image registration is an important research direction in robot-assistedsurgery visual navigation system. It will influence success or failure of the robot-assistedsurgery whether the problem of image registration can be resolved accurately and effectively,So the study of it has the important meaning. This paper mainly studies the human faceimage registration method which is used in robot-assisted surgery. Combine with thedemands of this research; it presents a method that based on marked points, which accordingto the principles of image registration.First of all, design marked points which are easy to extract, including theirs physicalproperties and position design. That can be applied to image enhancement by improving themethod of lateral inhibition network. Then combine with the color vector space and classicOTSU image segmentation method; make the color vector angle as a threshold to extract themarked points. In the marker matching, it proposes sub-block analysis method according tothe design of this paper, and matches each point sequentially by marking numbers.In the image registration, it puts forward the idea of applying coarse registration firstthen fine registration. According to the actual situation of image acquisition in thisexperiment, to do block image registration separately by using affine transformation method;fine-tune the image and make block image stitching by using rigid transform method. It isapplied fine registration on thin-plate spline principle, which makes the image distort like asheet. Finally, the floating image and reference image achieve complete registration.The inclination of the human face in the two original images is opposite. But theexperimental results indicate that after a coarse registration the floating image becomespositive face; then after a fine registration, the inclination angle are consistent with thereference image significantly. There are small errors by judging from the registration error ofthe marked points. In addition, the marked points which have characters of anti-disturbanceare easy to extract, also reduces the computation greatly. The results show that thereasonable validity of the method meets the experimental requirements in the system.