Due to the rapid growth of the Internet, there are tremendous amount of information (text, document, etc.) stored on the Internet. As a result, network search engine has been developed to search for the information by text. However, with the progress of software and hardware (Popularization of computer storage technology and digital camera in electronic device), it is easy to retrieve and store images and videos in the network(Picasa, Flickr, Panoramio, Youtube). In case someone is looking for pictures or information with specific object, scene or people, as shown in the figure above,the traditional full search applied to tremendous amount of information will result in an increasing of computation complexity. This full search technique is no longer appropriate.The problems above can be solved by Visual Search technique. The concept of Visual Search is transmitting the extracted feature set to database, and then searching the particular target from database. Finally, the matching results of the targets will be returned to the user. Moreover, the technique of Visual Search is widely applied to many different areas, such as the image search (Google Images, etc.), video search, 3D object search, and application of smart Television (second screen). Users are able to share the information (vote, lottery, etc.) searched from Television by visual search with people having portable device. Besides, users can search images by using portable devices. Furthermore, this technique can be combined with Virtual Reality to enhance the interaction between human and machine. In addition, MPEG is considering standardizing technologies of visual search applications. The background knowledge of Visual Search includes basic linear algebra, probability, image processing, computer vision, data mining, and machine learning, etc.

由於近十年來網路的興起,網路上儲存著我們無法吸收的資訊量(文字、文件等),於是有了網路搜尋引擎透過純文字搜尋(Text Search)。 然而,隨著軟硬體的快速進步(電腦儲存技術、照相功能電子產品大眾化),人們得以快速取得影像並在網路上儲存更多資訊量的圖片、影片(Picasa、Flickr、Panoramio、Youtube)。 假設人們想搜尋含有特定物體、場景或人物的圖片或資訊,如上圖情景,但網路上的資訊量過於龐大、類別也千萬種,最簡單的搜尋(個個比對)會導致搜尋速度太慢,已經不適用了。 VS的技術在於解決這難題,概念是對於影像擷取特徵點後傳至資料庫端,資料庫端的特殊搜尋演算法來尋找最符合的資料,最後再回傳給使用者得知比對結果 ; 再者,VS的應用非常廣,如影像搜尋(Google Images等)、影片搜尋、3D物件搜尋以及智慧型電視上的應用(第二螢幕),透過視覺搜尋電視上的內容來傳播有興趣的資訊(投票、抽獎等)給擁有手持裝置的收看者, 又或者是利用手持裝置影像搜尋等,也可更進一步增加虛擬實境技術讓人與應用互動。此外,目前MPEG組織也正在進行制定Visual Search的標準中。 VS所涵蓋的知識範圍包括基礎的線性代數、機率,到影像處理、電腦視覺、資料探勘、機器學習等。

Last Update 2014-03-02