| 摘要 |
PROBLEM TO BE SOLVED: To provide a method and a system for image search capable of accurately identifying corresponding points even in the case of comparing object images having many objects of similar shapes like building images having a large number of objects of similar shapes such as windows and pillars. SOLUTION: From a pair of a query image Iq in Fig.10(a) and a similar image candidate (street image) Iwc in Fig.10(b), two sets of corresponding point candidates [q1, w1] and [q2, w2] are extracted. In regard to one corresponding point candidate [q1, w1], the vertical position of the corresponding point q1 at the query image Iq side is located at a position N1 in an appearance pattern [N1-N2-N3...] while the corresponding point w1 on the similar image candidate Iwc side is located at a position N3. Since both the corresponding points have a large amount of deviation, they are discarded. In regard to the other corresponding point candidate [q2, w2], the vertical position of the corresponding point q1 on the query image Iq side is located at a position N2 in the appearance pattern [N1-N2-N3...], and the corresponding point w2 on the similar image candidate Iwc is similarly located at the position N2. Because both the corresponding points are close in distance, they are determined to be corresponding points. COPYRIGHT: (C)2011,JPO&INPIT |
