Optical flow in an image is produced by motion. This technique doesn't require preliminary feature detection. Motion rather than shape is computed [152][90], a low-level processing of the pixel data is performed. It is a direct representation of facial action after proper smoothing is performed on the motion field. The optical flow based method is quite sensitive to noise and cannot capture rapid motion. However, the method of Essa and Pentland [42][44], which employs a multi-grid optical flow coupled with a physical model of a face appears promising, and captures spatial and temporal variation of facial motion in sufficient detail.