1996 年 16 巻 3 号 p. 138-149
Stereoradiographs have features of high spatial and high temporal resolution as an examination for blood vessels, and give us useful internal geometrical information. However, viewing point is fixed to the point of X ray source, so that radiologists perceive depth with the technique of “cross eyed stereoscopy” in general. We are studying 3D reconstruction using a pair of stereoradiographs in order to observe from any direction and obtain geometrical information. In this paper, we described a method of geometrical determination and an edge feature extracting procedure for finding correct binocular correspondence. The edge feature extracting procedure is modeled the procedure to find binocular correspondence on the visual nervous system. The procedure can be classified into 2 main processes.1) The contrast enhancement process; original images are convolved with difference of two Gaussians (DOG) operator. The DOG operator has a characteristic of a small central excitatory part and a larger inhibitory region that extends over the entire 11 by 11 receptive field.2)The edge features extracting process; the contrast enhanced images are convolved with operators that have a sensitivity of specified direction. Using these results as excitatory input and inhibitory input over the receptive field, direction and position of edge are recognized by processing with the “shunt” characteristic. We also investigated relation between spatial resolution and pixel size. Finally, we presented examples of automatic 3D reconstruction using stereoradiography.