Y-Vid | V_A

Y-Vid

Y-Vid is an application specially designed to perform automatic video summarization. It provides as output video summaries of long feature films of various genres, comedies, detectives, war films, etc. A video summary represents the content of the film so that the spectators don't need to see the whole film thus saving much time. It will take 56 minutes to watch "Gone with the Wind" instead of four hours, and 20 minutes to watch "Ruthless People", for example. The user can save lots of his/her precious time.

The generalized algorithm underlying the application's functioning includes the following steps.

- Convert the film into text.

- Summarize the text using traditional summarization techniques, for example symmetric summarization (see [32] in the Publications section).

- Extract the most important information from the text and map it on film episodes.

- Extract the corresponding film episodes and merge them into video summary.

The resulting video summaries are usually 3-4 times shorter than the original film.

The application's input will be a movie loaded by user. Its output is a video summary of that movie. The application will support all most popular formats of video files. At the initial stage it will support processing movies in English and Russian. Then it will be adapted and localized to process all languages that use Latin and Cyrillic alphabets.

The amount of money needed for implementation of the project is $100,000.

The money will be required for

1) Creation of video summaries by human experts.

2) Creation of the manually annotated corpus of scripts of summaries and original movies.

3) Development of machine learning algorithms.

4) Creation of the application.

V.A. Yatsko's team has profound skills in NLP and all range of probabilistic techniques including Bayesian models, Markov chains, odds ratio, information gain, etc (see Publications). We have applied machine learning algorithms since 2010 (see Products and Downloads) . V.A. Yatsko has been engaged in solving summarization problems since late 1980s.

See sample video summaries at VK and here.