3D Remote sensing images online refining

Abstract

A depth perception in binocular vision is based on horizontal parallax. However, if horizontal parallax is outside a certain range, for example, if parallax is larger than eye separation, then diplopia occurs and the 3D depth illusion collapses. Errors may be produced during the procedure of automatic image matching, epipolar image resampling and color 3D anaglyph image generating. Moreover, because satellite images cover a large area, some areas with high elevation have larger horizontal parallaxes. Therefore, there may have excessive horizontal parallax between some correspondence points (homologous points). In order to reduce image fusion problems, authors present online refining algorithm for 3D remote sensing images. Firstly, build the pyramid of parallax information file for all correspondence match points according to data organization of 3D remote sensing images. Each file is with respect to each layer of image pyramid. When refining, compute the maximum parallax and the minimum parallax in the display window area. Secondly, compute the pixel number of shifting image according to the maximum parallax and the minimum parallax. Thirdly, shift red channel image of right image in epipolar image pair. Fourthly, merge blue channel and green channel of left image with red channel image of shifted right image to produce new refined 3D images. At last, return the images to browser side. With the proposed method, we can reduce overall maximum horizontal parallax and maintain an acceptable parallax range for 3D viewing. The system can make user automatically or manually adjust the parallax of 3D remote sensing images at the same time of browsing 3D remote sensing images on Internet.