A New Nonlinear Distortion Correction Approach for Camera-Projector System

-This paper presents a new approach allowing a projector to display an undistorted image on an unknown geometry surface. In the distortion correction system, a single digital camera is used to capture the viewer’s perspective of the projection surface, and a computer is used to process the origional source image to show an undistortion image from the view position of audience. In the approach, no explicit camera and projector calibrations are needed. The procedure establishes the mapping from camera image to projector image by a well trained neural network (NN). After the mapping has been establised, a transform converter table is constructed from the output of NN. With the transform converter table (TCT), image correction process can be realized in real-time, therefore, this approach is avaiable for various applications, such as movies and etc. By using of NN, we do not need to calculate all image pixels for establishing the mapping transformation. The mapping can be obtained correctly by less computation if enough teaching points are provided. The simulation results show that our approach is effective for correcting distortion caused by continuous projective surface and practical enough for most projector presentations. Key-Words:-Projector-camera system, Projective mapping, Distortion correction, Neural network, Computer graphics, Keystone