Lip motion has been used in speech separation to enhance the speech signal of interest or directly in speech recognition to improve the recognition accuracy. This thesis presents new techniques in face detection and lip feature extraction, both of which are prerequisite to the use of lip motion. In the face detection technique, the fact that faces have strong edges around the facial features is exploited to significantly reduce the model's size to 5.06Kbytes, yet achieving the detection rate of 87.2% under various lighting conditions. This allows us to build a compact hardware system. The lip feature extraction technique uses the contrast around the lip contour to extract the height and width of the mouth - two important parameters in speech separation. An FPGA implementation is presented, that occupies 15,050 logic cells, or about 6 times less than the current leading FPGA face detection system.