Abstract
An efficient algorithm for searching similar images is proposed. The algorithm is applicable to content based image retrieval and multiple template matching. For designing the algorithm, the upper land lower limits of the correlation coefficients between registered images and a object image by an indirect way through a pibot image. An interval estimation based on the limits shows the range of an unknown but estimated correlation coefficient. Using the interval estimatin, some conditions for selecting redundant images of smaller similarity from the registered images can be detected and skipped away from conmputation, which enable to effeciently save computational cost of searching. The algorithm is applied to real searching problems in a image database of 1,200 images and in multiple template matching problems, showing the effectiveness of the proposed method.
