Semi-automatic annotation system for home videos

Abstract

This paper presents a semi-automatic system for home video annotation that searches into the video contents and retrieves video shots for a specific person. The proposed system is composed of four phases; 1) shot detection phase that detects shots boundaries and divides the original video into shots, 2) face detection and recognition phase that detects faces in video shots based on Haar-like features and uses the Principal Component Analysis (PCA) algorithm for features extraction in order to select the eigenvectors with the largest eigenvalues, 3) face clustering and annotation phase that groups faces with similar facial features into the same cluster and apply face annotation on persons' shots (person's faces) using User Interface, and 4) retrieving phase that enables the user to enter a query to search by person's name then retrieves the video shots for this query from the database of the video shots. The proposed system is simple and provides a friendly user interface. It greatly reduces workload and enhances the accuracy of annotating person's faces in home videos.