STUDY OF THE 2D PICTORIAL CUES ON NATURAL IMAGE'S STEREOSCOPIC FUSION
 
Yuan, J.C., Chern, Y.J., Fu-Jen University, Yeh, Y.Y., National Taiwan University, Huang, K.C., Hsueh, W.J. and Tsai, C.H., Industrial Technology Research Institute Taiwan
 
From the demonstration of Julesz's random dot stereogram, we know that without monocular cues people can see the 3D from. Now from the applied point of view, concerning the 3D images, the personal computer and visual display terminal (VDT) are so popular, so the presentation techniques of stereoscopic images are becoming much easier. Stereoscopic television might be expected now, and which always present the natural images. So, we do some experiments to collect the human factor data about the 2D pictorial cues effect on 3D stereoscopic fusion. The first experiment we let subject to evaluate the goodness of pictorial depth cues on each images (Gestalt) and the number of pictorial depth cues (eg. Interposition, shadowing, perspective, texture cues) in each pictures. Then we measure the correctness and goodness of the stereoscopic fusion of each pictures (positive disparity, zero disparity, and wrong disparity). The result shows that the more and the better of the 2D pictorial depth cues, the better and the more correctness the stereoscopic fusion. On the second experiment we manipulate the luminance range of the pictures. We divided the pictures into two groups, one has good 2D pictorial depth cues, and the other has not. The results are the pictures have good 2D pictorial cues has better tolerance on luminance compress than the bad 2D cues group. At the same time, the natural images compress their luminance to lighter range (eg. 100 to 255) will brake down their stereoscopic fusion more seriously than their luminance compress to darker range (eg. 0 to 155). If the pictures cover the full range of luminance (0 to 255) and have no sharp local contrast, then the stereoscopic fusion will be the best. These results will help people to make a better film or pictures under proper luminance condition, and also offer a data for the (video) camera company to design their instruments. The third experiment simulates the gradient of ghost images. Because the superimposed images of left and right eye have the crosstalk problem, so, we depicted the degradation curve. The results show that the better the monocular pictorial cues, the more the tolerance of the crosstalk. Keywords: Random dot stereogram, crosstalk problem