Browse Prior Art Database

A method to filter video scenes of interest based on role face recognition Disclosure Number: IPCOM000239491D
Publication Date: 2014-Nov-12
Document File: 5 page(s) / 114K

Publishing Venue

The Prior Art Database


This invention solves the problem that video content cannot be filtered based on user customized parameter. This invention provides an overview of characters in the video. All human faces are identified in the video before it's played. The video content is visually summarized by displaying the identified images of characters faces in order on the screen based on the total time length of occurrence. Then, users can filter wanted characters in the video by selecting the image(s) of role(s) who they want to see in the video. Scenes that do not contain the selected images are greyed out and abandoned. The video length is cut to play only the scenes of interest.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 62% of the total text.

Page 01 of 5

A method to filter video scenes of interest based on role face recognition

When watching a video that involves people performance, such as movies, TV series, or entertainment programs, people usually have roles or actors that they particularly like. These roles appear in different scenes and at different time as the video plays. If one just wants to see the scenes that a specific actor appears, he or she has to drag the progress bar, preview the video to navigate to the relevant scenes and filter out irrelevant scenes. Imagine the following scenario:

A US drama you're following consists of 100 episodes and each episode is of 40 minutes. You want to look at scenes about a specific role, but for each episode, you will need to manually locate. When you start watching the video, you only find that in this episode, your interested role only appears for 10 minutes but before that you do not know. You wasted 30 minutes on scenes you're not interested. Time is precious. If there is a way to help filter out the information or scene you like, that would be very helpful.


When the user triggers the role face filtering mode, before the video starts, it can automatically identify all of the human faces in the video, calculate the time of occurrence, and the total time length each role appears according to the time length of the identified face. The result (face image and the total occurrence time) is displayed as a graph in the sequence decided by the total occur...