An initiative of NATIONAL INTERNET EXCHANGE OF INDIA & DIGITAL EMPOWERMENT FOUNDATION

Hozaa – From Pixels to Searchable Content

More projects on :

Description

Hozaa is a breakthrough in video search engines, making large video repositories searchable by recognizing unlimited number of objects, concepts, and events. Conventionally, tags are added to videos by users, and these videos can be browsed or watched if the search matches the tags associated with them. Hozaa has gone a step further by using artificial intelligence and computer vision to interpret and understand video data, and assort videos with respect to their content. It then generates appropriate ‘content tags’ for recognized objects, concepts, and events. As a result when user types a text query, Hozaa uses these ‘content tags’, in addition to user tags, to retrieve a much broader set of precise results as compared to traditional search. In addition, Hozaa revolutionizes the way users search for videos by offering reverse video search capabilities. Now users cannot only perform more accurate video searches with a display of a wider range of similar videos but can also search for videos by uploading related images.

What we deliver

1. For ordinary users, the benefit lies in the accuracy of the results which saves them time they have to spend to get to such results. Not to mention all those videos that they could not reach altogether because of poor/in-accurate or no tagging. 2. The same applies to copyright folks who may/may not be able to detect certain breaches because the tags have been used intelligently by users to go around them. Also imagine the situation where violation takes place in a different culture and tags are specific to it, unless one looks at the content there is no easy way to identify this kind of breach.

Whom we deliver

First of all users have choices, they can use three kind of searches: a. text b. video c. image What’s more all these searches work in combined way, the users can switch between them and also use results from one search to kick-start the other. This way the tools at user’s disposal make the job of searching for the right video content simple, quick and effective.

Why is the project unique?

Recent efforts in improving content aware video search and retrieval have been limited to face recognition and concept detection. Video search engines such as VideoSurf and Viewdle have the ability to recognize familiar faces in a video and tag them accordingly. Concept detectors, developed by Columbia University, University of Amsterdam and IBM, rely on a very limited set of concepts such as face detector, car detector, fire detector etc for video content extraction. This approach requires precisely labeled video or image data for training purposes, and proves to be tedious and expensive on the internet scale. As opposed to the above mentioned approaches, our proposed approach for video search and retrieval is completely data driven, and therefore, is not to be limited to ‘faces’ or few hundred concepts. The gist of our approach is to employ noisy tags assigned to millions of videos by internet users in conjunction with content analysis algorithms and generate more specific ‘tags’ content oriented tags that truly represent the content of a video and its various interpretations. In other words, by combining noisy tags with intelligent content analysis, we convert the visual information of a video into meaningful searchable text or meta-data. Our approach is scalable to video repositories of any size, and due to its data driven nature the search performance and accuracy increases with the size of the video repository. Hozaa has revolutionized the concept of video searching by adding the all new image to video search feature. This gives users a unique experience by letting them upload images from the internet on the Hozaa website and retrieve videos associated with that image.

Contact

Babar Khan

TechArete (Private) Limited

3rd Floor, Software Technology Park 1, F 5 / 1,

Islamabad, Federal Capital, Pakistan

44000

URL/Website – www.techArete.com

Submit a Comment

Your email address will not be published. Required fields are marked *