Interframe image coding with 3-dimensional edge detection

A new predictor has been developed for interframe image coding. This predictor uses a three-dimensional edge detection technique to predict the present pixel from the three previous frames. The assumption in this research is that the motion is approximately linear within a few frames in the normal videophone sequences. The three-dimensional (spatio-temporal) gradient vector is estimated, using three-dimensional edge operator masks; the tangent plane to this vector is then calculated and projected to the current frame, which is two time units behind the center pixel of the 3-d edge masks. The prediction pixel is found in the previous frame along the intersection with the tangent plane. The masks in the spatial domain are enlarged when the prediction value is not close to the center pixel two frames behind. The comparison with the Netravali pixel recursion method shows that the mean square errors are lower at similar bit rates in most of the frames tested