| 摘要 |
A method of and a system for finding similarities between major boundaries of images using a wavelet detector is described herein. Unimportant edges of the image are disregarded by eliminating Gaussian wavelet coefficients and Haar wavelet coefficients of lower significance. Comparison between the images is made on the basis of quantized color, sign and magnitude of the Haar wavelet coefficients. The method performs the comparison between images in two steps. First, the method checks for exact matches between the Haar wavelet coefficients to determine whether the images are very similar. This is followed by binning of the coefficients into nine spatial bins in the image. A representative is assigned to each of the bins in terms of color, orientation and sign. Each bin of one image is compared with all the bins of the other image. Thus, images that are similar but not identical are still detected.
|
