There are specific algorithms for image matching. One of the more famous ones is to use Wavelet coefficients. Wavelet Coefficients are resolution independent ways of representing image content (The image data is mapped into a 0-1 2D space). You can perform a 2D difference analysis of the searched image's wavelet representation with the Wavelet representation of the images in your index and then rank the results based on the degree of matching pruning the results based on thresholds. The challenge here is to scale the algorithm to the scale of the Google Index data. I'll leave that discussion for another thread.
(For more info start here on wikipedia: http://en.wikipedia.org/wiki/Wavelet_transforms). Note: a course in Image Processing is helpful. Also, this method works even if the image is resized. (Side Note: Rotational changes don't affect the first few coefficients, but affect the later ones) ^ This answer is only for the "Search by Images" feature. For the image search by text features, you use the standard Google way of associating images with their nearby textual content and then search based on those tags. -- DK http://twitter.com/divyekapoor http://www.divye.in -- You received this message because you are subscribed to the Google Groups "Algorithm Geeks" group. To view this discussion on the web visit https://groups.google.com/d/msg/algogeeks/-/3a_k2ON4CJ0J. To post to this group, send email to algogeeks@googlegroups.com. To unsubscribe from this group, send email to algogeeks+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/algogeeks?hl=en.
