Hi Kim, On 7/24/07, kim wroblewski <[EMAIL PROTECTED]> wrote: > I am working on an image filter that puts textual descriptions of images > into the content attribute. For example it will tell you what color the > image is. > > A search for "Brown" will return only a single image, while there are in > fact many images with Brown as its content. Or for example "SeaGreen" > returns no images at all, while I know there are images with that color. > > Can anyone shed light on Lucene's logic here?
The logic is actually Beagle's. It stores in its database the metadata however it is extracted by the image filters. I would suggest running the beagle-extract-content tool on the images to see how the properties are set in the index. In Beagle fields generally are either considered "text", in which case the full text is tokenized and analyzed and searches match it, or "keyword", in which the exact text must match. For example, a text field with the content: the quick brown fox jumped over the lazy dog would match against searches for "brown", "jump", "jumping", "dogs", etc. However, a keyword field with the same content would have to match the entire string to be displayed. If I had to guess, the images you're describing are probably all being set as a single keyword. Joe _______________________________________________ Dashboard-hackers mailing list [email protected] http://mail.gnome.org/mailman/listinfo/dashboard-hackers
