| Summary: | The increasing availability of high-speed internet connections, the increase in smartphone usability and also the ubiquity of social networking, all combined, help to create a great diversity of User-Generated Content (UGC). Along with this expansion, Ultra High Definition (UHD) broadcast technology has been developing rapidly since its beginning. This created the need to distinguish between good and bad quality videos. The best way to assess the quality of a video is through the human eye. However, given the amount of content it becomes quite impractical. Therefore, computational methods are used. These methods try to assess it as close as possible to what would be assessed by the human vision. The semantics of a video is the meaning of the video itself and using this information, an idea of what the video is about can be provided, helping even in the assessment of a video. Having that in mind, this thesis uses a video collection and a news articles collection in order to extract the information regarding the objects in the scene and the terms in the news. The similarity between both information is taken into consideration to assess the quality o the videos. In this way, the assessment is done using semantic information. The main contributions of this work are the video quality assessment based on semantic information and an evaluation of a set of object detection algorithms used for semantic extraction in videos.
|
|---|