Face features detection and representation by morphological reconstruction and linking

In this paper, we try to address the low level image features detection and representation problem under various illuminations and in various backgrounds. We use multiple thresholds iteratively get a proper threshold value. The results of detection may have disconnected lines, or edges, or boundaries. We use morphological reconstruction and edge linking algorithm to rebuild the edge and connect the disconnected boundaries. Also, we use a set of straight line to approximate the contours or boundaries of the face features, which can simplify the representation of these features. Experiments on human face images show that the results are good.