| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To solve the problem that it is difficult to determine vanishing points with respect to various images automatically, and that it is not easy to form correct depth structure models automatically and realize three-dimensional vision without a sense of incompatibility, even when perspective structure presumption suits. <P>SOLUTION: To prevent a person from having a sense of incompatibility with respect to any kinds of images, and moreover determine a scene structure as close to a reality as possible, three kinds of fundamental depth models representing depth values concerning individuals of three kinds of scene structures to be fundamentals are stored, in frame memories 4, 5, 6. A composition ratio 8 is determined according to high-frequency component evaluation values of the brightness signal of an input non-three-dimensional image from a high-frequency component evaluation section 2 of an upper part and a high-frequency component evaluation section 3 of the lower part, and according to the composition ratio 8 three kinds of fundamental depth models are compounded. The compounded fundamental depth model is superposed on the R signal 9 of the non-three-dimensional image in an adder 10, to obtain final depth presumption data 11. <P>COPYRIGHT: (C)2005,JPO&NCIPI</p> |
