Transformation of Multi-camera Images for Free Viewpoint Television

Abstract

This paper introduces a preprocessing method for Free Viewpoint Television (FTV). In every multi-camera system, it is difficult to precisely align cameras in the desired place. For FTV, this misalignment reduces the efficiency of image compression and the quality of interpolated images. We propose a preprocessing method that transforms the captured multi-camera images by using projective transformation matrices to compensate for this misalignment. This transformation is able to improve them because the multi-camera images become more correlative. Furthermore, we introduce a method for automatically finding the projective transformation matrices using the epipolar geometry. Experimental results show that PSNR gains of up to 3.0 dB in comparison with compressions without any processing, when MPEG-2 is used under conditions ranging from 1.6 to 3.0 Mbps. The quality of the interpolated images is also apparently improved.