This paper describes a fast algorithm that tracks face region in dynamic images. We adopt the color histogram backprojection algorithm of swain-Ballard in order to find face region in subsequent frames. The algorithm has the advantages of being invariant to object rotation, small deformation, and noise. The following problems are addressed: building a model histogram, backprojecting model histogram, searching the peak response location, and updating the model histogram. The experimentation shows the possibility of realtime face tracking on PC or low-performance workstation.
"Method for real-time face region tracking based on color histogram", Proc. SPIE 2739, Acquisition, Tracking, and Pointing X, (7 June 1996); doi: 10.1117/12.241932; https://doi.org/10.1117/12.241932