Reverse image Search and mobile e-commerce

Yinzidu

For some rainy afternoon, he sits alone in a café in the city, a crisp ring, push the door and into a carrying a board of MM, Qing Qing can not square, he felt déjà vu but can't think of where he saw, then quietly picked up the phone took a clear picture, import an app, a second later, The MM details are shown, he found that the original is 10 years old classmate Han Meimei, then walked over ...

"Long time no See ..."
"Are you?" ”
"Li Lei. ”
Lighting, music, depth, YY
......

Traditional image search, the implementation is through the description of keywords (descriptive text) to match the image keyword (with a higher degree of text associated with the picture), that is, the existing image search function (such as Google, Baidu's image search) in essence or text search. However, the object of Li Lei in the above story is the use of "picture source" search information (including pictures and text), so exciting technology, what will be the application of the future.

Let's look at some interesting products.

1, RETRIEVR

Flickr's old users are certainly not unfamiliar with RETRIEVR, which is one of the earlier sites in the industry to provide image search image services.

RETRIEVR provides two ways to search, one is graffiti search, users can use a small graffiti editor to draw the outline of the image (a bit like a vague search), and another way is the image upload (local and URL two modes). In short, the search target is compared to the image database on Flickr's main server, and the approximate graph is found according to the matching rules. In fact, RETRIEVR's recognition is not high, and only for the photos in Flickr matching filtering, is not practical, is now more suitable as the Flickr platform for leisure development tools.

2, TinEye

TinEye is much more reliable than the RETRIEVR search function. It is an image search site developed by Idée Inc., the parent company that designed the very NB Pixid. It provides the same search methods as RETRIEVR, uploading local images or providing URLs for online images. TinEye also provides plugins and Favorites script buttons for IE, Chrome, Firefox and other browsers, and users can search for similar images in the browser using the right button menu.

Official data show that as of July 19, 2010, TinEye's database picture has reached about 1.63 billion, users can even search the same type of map (such as magazines) through a source map.

Reverse Image search Engine

This image as a search source of technology can be called "Reverse image Search engine", the most representative of Google's "scene search" Pilot product Goggles (it can be uploaded to the user's image content analysis, can identify books, music album cover, painting, Landmarks, sites and corporate logo patterns, etc.).

"Our ultimate goal is to identify any image," said Vick Gudera, Google's vice-president for engineering technology. This technology is our initial research in the field of computer vision (computer vision refers to the use of cameras and computers in place of human eyes to identify, track and measure targets). You can take a picture of the scene, and then use the photos you took to search the Web.

Reverse image search engine use scene is very complex, here first List 3:

1, "to look for the map", this situation is most common, such as we see on the Internet a very like the picture, unfortunately the size is too small or the quality of the picture itself is not high, this time you can upload small map files to search for large high-definition image.

2, through the local image search whole map, for example, the archaeological more interested friends can according to a piece of cultural relics "proofreading" to search the whole piece.

3. With the development of mobile electronic commerce, the demand of image search service is becoming more and more urgent. Just imagine, in the shopping when suddenly found a very interesting little things, casually photographed, uploaded to Taobao, suddenly appear gorgeous list ...

Market trends

In the face of such a tempting market, it bosses are certainly not idle.

After Google released Goggles in 2009, Bing invested its own image-search technology, and Google bought Plink and like.com.

Plink: has developed a mobile version of the image search engine Plinkart, which is based on the Android mobile platform application, it can help you search for your mobile phone camera pictures of the art information.

Like.com: Formerly a company called Riya to develop facial recognition technology.

Domestic, Alibaba Group Investment image Search website "Amoy Search" has been on line, currently only for some women's category to search, Baidu's new product "Map Search" (currently in the beta phase: shitu.baidu.com), is about to be listed.

Technical limitations

Once and a picture search field experts chat, the feedback is based on the current technology, but also limited to search for rigid objects (such as buildings, books, etc.), and material, color, relative coordinates of anything is not reliable. In addition, the image search ≠ image retrieval, only when the available data reached millions, called "search", so, mobile internet application domain "picture search" (such as mobile phone photo search function) The biggest bottleneck is not "billion" as the equivalent of available picture data.

In other words, the technical limitations of the reverse image search is that there is no uniform matching standard for the entire Internet cloud Picture database, which results in most image search products serving only their own data sources, playing small audiences and not being strong in expansion.

Image Search and mobile e-commerce

With the take-off of the mobile Internet, E-commerce based on mobile platform will emerge in a new industry blowout in the next few years. In fact, people's exploration of the new mobile shopping experience has never ceased.

such as Shazam Encore based on audio recognition technology, you can input existing music, parse and search the audio source file, and finally connect to the music website to complete the purchase. --auditory-guided buying experience

Another example is TinEye Music, which uses the iphone to capture all the information about the CD in itunes, allmusic.com, YouTube, and Wikipedia with the CD cover. You can listen to the album, review the album reviews, or buy the CD directly from the search results. --Shopping experience of visual purchase

Therefore, as long as the corresponding shopping platform has the ability to prop up all product class purpose picture classification, and can form a relatively complete picture matching rule, this piece of commercial market can use "infinite" to describe--this will be to the traditional shopping experience revolutionary subversion.

Fragmentary thinking (thanks to the students to share their ideas)

1, before the exact match, reverse image search engine should be used with auxiliary filter control, that is, upload pictures, you can also enter other filtering information, improve the accuracy of search results.

2, in addition to the complex filter control, you can also improve the accuracy of the search results through more perfect recommendation system and matching rules. (Regarding the recommendation system, will be discussed in detail in later articles)

Source: http://mdchina.org/?p=184