Detection and removal of adherent noises in video from a moving camera

A novel method to detect and remove adherent noises in videos from a moving camera is presented in this paper. The basic idea is to detect the adherent noises by spatio-temporal image processing technology first and then remove and restore the information in noise regions using a 3D inpainting technology. Under the condition that camera motion is unknown and unconstrained, a 3D spatio-temporal image is acquired by a perspective transformation without motion estimation and the static background of the spatio-temporal image is modeled. Then the adherent noises are extracted by tracing trajectories of the difference between adherent noises and static background and the data is eliminated in regions of adherent noises. Finally, a 3D spatio-temporal exemplar-based texture synthesis approach is applied to inpaint the miss information. The proposed method is shown to be very effective on real video acquired from a moving camera.