Abstract
This study proposes a binocular vision model that reconstructs 3D structures from autostereograms. Our model is a modified version of the model by Hayashi et al., which cannot resolve the autostereograms, and serves to introduce fusion processes by vergence eye movements in order to obtain depth information from autostereograms. Since autostereograms have a periodic plane as the background, our model starts with the vergence motion of two windows corresponding to the left and right eyes to search maximal output of the binocular energy unit that produces autocorrelation-like patterns of the background. Once the optimum vergence angle has been determined, the model calculates paired or unpaired points using the same procedure employed in the model by Hayashi et al. The results indicate that the model can reconstruct 3D structures even from autostereograms, suggesting that the disparity detection system coordinates with the eye movement system to accurately derive depth perception.
