View-Based and Visual-Attention-Based Background Modeling for Detecting Frequently and Infrequently Moving Objects for Video Summarization

Abstract

The real-time face detection (FD) algorithm is proposed to find faces in the images as well as videos. Besides face regions, this algorithm also finds the exact localities of the face parts like lips and eyes. Initially, skin pixels are extracted centered on the rules of simple quadratic polynomial model. By introducing small modifications, this polynomial model (PM) could be applied for extracting the lips. The merits of adopting these two identical PMs are two-fold. Firstly, computation time is saved. Secondly, these extraction processes could be executed at the same time on one scan of the video or image frame. Subsequent to skin and lips, the eyes are extorted. Later, the algorithm eliminates the falsely extorted parts by validating with rules taken as of the spatial and geometrical relationships (SGR) of face parts. At last, the exact face regions are ascertained accordingly. As per the experiential outcomes, the proposed algorithm evinces preeminent task in respect of accuracy and speed for FD with huge differences in color, size, shape, expressions, and angles.