Color and heuristic-based face detection in H.264 video sequences

We presents a color and heuristic-based face detection algorithm in order to detect faces in H.264/AVC compressed domain. Face detector is composed of preprocessing, selecting face candidate and face authentication. Firstly, video capturing conditions are removed by the modified Successive Mean Quantization Transform (m-SMQT). Secondly, face candidates are run by using skin color image segmentation, H.264/AVC DCT coefficients and well-known face features. Finally, face locations are confirmed by using frontal face template and geometric structure. In order to achieve correct and speed requirements, we introduced m-SMQT, high-speed rectangle region partitioning algorithm, 16 points up-sampled AC image and K-parameter-based face feature clustering technique. The proposed algorithm is not only affected less from the different of face poses, but also offers low computational cost solution to problem of detecting faces. The face detector achieved a high recall value of 81% in detecting faces of sizes more than 14×12 pixels in all I-frames of H.264/AVC QCIF video sequences.