| Čoahkkáigeassu: | This report addresses the problem of recovering the person-of-interest in a broadcast video. This is extremely challenging due to the imaged appearance variation of the person-of-interest in the video. However, the abstraction of the meaningful data is somehow similar to a table of content; it can serve as a useful index table, enabling users to browse through huge amounts of data in a non-linear manner.
In order to achieve a useful index table, key frames are commonly used in applications such as video editing and non-linear data browsing. This report presents two methods for key frame selection: capturing of Euclidean distances between their color histograms and secondly, measuring the similarity between two images by viewing the images being compared with the other image regarded as perfect quality. Thus, this report will present two discriminative key-frame selection methods, and the experimental results will be discussed.
Lastly, based on the sparsity, a recognition performance can be achieved by a simple algorithm and the face of interest will have the name label on it.
|
|---|