If you’ve already played or worked with xcavator.net, you may have become curious about how it works. Here is a brief description of its inner mechanics.
Xcavator.net provides several tools for finding the right image. You may:
1) Narrow search results with images by using keywords;
2) Select images with a certain dominant color;
3) Find images that look similar to a selected reference image, and finally;
4) Interactively choose certain parts in the reference image to make them and their spatial relationship more prominent when searching for similar images - the 'traces' function.
In our solution, all these tools are implemented with just two distinct search engines: one for keywords, dominant color and image similarity and the other for the 'traces' interactive visual search. The former one is based on so-called indexing techniques, so I call it index search, which combines visual and text information into a single data structure and produces results very quickly, regardless of the size of an image collection. The results of our indexed search are superb for this class of algorithms, and we can handle significantly larger image collections than our competitors.
The purpose of the second interactive search engine is to further tune the search results to specific user needs. Interactive search is much more accurate and specific than the index search because the user can select important image parts and skip unimportant ones. We call this 'traces' on the site because the user can pick points of interest or trace through a feature or feature using the mouse. The interactive search is applied to the results of the index search as a refinement tool, and it brings the best images to the top of the search results based on this additional user-defined search criteria.
There is a lot of interesting things to be said about the interactive search. I’ll cover this topic in one of my next posts. Meanwhile, I highly recommend watching our Video Intro to learn how to use our search engines in the most efficient manner.
Labels: image search, indexing, photo search, search engine, search results, stock photograpy