Real-Time Human Object Extraction Method for Mobile Systems Based on Color Space Segmentation

Abstract

Since a human object is an important element of the moving pictures being processed by mobile terminals, establishing a human object extraction method encourages dissemination of new applications. In accordance with the requirement of mobile applications, this paper proposes a low-cost human object extraction method, which consists of a face object and a hair object extraction based on their color information and a simple body extraction utilizing the position information of the face object. In the proposed method, skin color and hair color are estimated through color space segmentation, and a human object is effectively extracted by using a radial active contour model. Simulation results of the human object extraction with the use of XScale processor claims that QCIF 15 fps video sequences can be processed in real time.

Real-time automatic moving object extraction and tracing based on improved active contour model

“…It can be classified into two catalogs, automatic and semi-automatic, according to the extent of user interface. Many automatic video object extraction algorithms have been proposed [1], [3], [4], while semi-automatic methods [2], [5], [6], introduce human interface to define the semantic object. Different methods were developed for video object extraction.…”

“…Homologous region grow and cluster method [1], [7] is usually time consuming so that real-time application is difficult. Contour based method [2], [4], [5], [6] is fast and can detect exact boundaries, but generally needs user interface and sensitive to noise. Restriction of motion based method [2], [3], is that only motion is not sufficient to define semantic objects.…”