Image search of reception of Cabinet

How a pile of lousy image search results out clue? clustering image search will bring different experiences.

The Internet weekly Yang-reporter

Now the market picture search engine a bit lazy wardrobe.

When you use the keywords search picture, often because the keywords themselves simplistic or fuzzy, lead search results disorganized. In the face of search engines for thousands of results and impatiently turning down, your mood will certainly like the Pro before going out in a tidy raking of the wardrobe to wear clothes.

Have a professional do a survey comparing search pictures of experienced and limited experience, the two groups of people and found a common rule: when they search for pictures, often does not take the initiative to change keywords, but in the result back in until you find satisfying.

Don't blame them for being lazy, this move is because, in the search for a picture, a lot of people do not know how to define the keywords and columns too much and often counterproductive.

(In the IGroup "search type" Tiger ", will automatically generate a lot of" Tiger "keywords related to the pictures folder, including golf star" TigerWoods "picture clips)

New technology can bring about change? from a Microsoft partner, launched a "named" IGroup (ImageGroup) picture search clustering in the direction of the research project, have a good solution to users of these troubles.

Like K song, love the music of the wind, and researcher at the King was dressed in a stylish, full of affinity for user experience designer wangshuo is definitely two of Microsoft's personality, which is responsible for the project-specific algorithm and technology, the latter from the user experience perspective design entire interactive processes and interfaces.

Their IGroup and their own, are a bit beyond traditional approaches, open a new taste. Currently, IGroup in certain key technology has been patented (http://igroup.msra.cn).

IGroup can schema in an underlying search engine.

It took users of the search interface is divided into two zones, the left is related to the keyword phrase clustering and small thumbnail on the right is the picture search result display area. This seems a bit like a classification of the wardrobe, and check all the categories are automatically completed. For example, when you search for "Tiger" (Tiger) is a term, the left side of the cluster will show related several phrases, in addition to be broken down into "Bengal tiger, India Tiger" and "Siberian Tiger" category will appear similar to the "TigerWoods" ("Tiger" Woods, a famous golf players) in the category, and even Apple had release, code-named "Tigers" of the operating system. The user can be the first on the left under the classification entries and small thumbnail for the primaries, then in the related right display area in detail.

In the browse area to the right, the initial results have multiple display arrangement.

The first is from the left navigation area of each category in a few representative picture of clustering thumbnails; the second is to automatically display a larger image clustering.

This combines the semantics of inductive view can significantly improve the efficiency of the picture search and can give the user additional tips.

For example, when you search for "the pentagon" (Pentagon), IGroup in navigation area lists the United States the Pentagon and pentagonal (pentagonshape) different clustering phrases for users to choose, so the user can in the next query, direct input once visited by a term as a keyword, and thus obtain more accurate search results.

In fact, about picture search clustering is not a new topic.

Traditionally, on each picture to extract a feature (text or image), and clustering. But this approach is evident in the efficiency and speed at a disadvantage. IGroup has selected a clever approach based on the user's picture search keyword, and then in the search engine text search, and extract out before hundreds of result in recurring high frequency phrase as the second image search keywords (around 20), and for some words that cannot be classified is deposited in an "Other" (other) category, which covers all of the search results, without affecting the speed of the increased efficiency of the query.

In the algorithm, how to define and obtain accurate cluster, how to eliminate the redundancy of the various types of pictures and so on, are the difficulties of IGroup.

King Air, and also try wangshuo further optimization of the project. In the future, they may take a similar mechanism, under digg.com users search and click on clustering and sort thumbnails; or may be present in the cluster are divided into several levels, so that users can further refine the selection.

A variety of ways, but it can be expected as a result, future image search will become more and more intelligent, more and more convenient.

If you want to experience ImageGroup technology, please visit: http://igroup.msr

a.cn