Detection of a change of scene in a motion estimator of a video encoder

A method of motion estimation from successive picture fields for a video picture coding system comprising decoding of pictures wherefrom temporal redundance is removed through a motion estimation carried out by macroblocks of pixels in which the pictures are divided, the method comprising the steps of:

• defining and calculating a smoothness index of a motion field of each picture by analyzing motion vectors of all the macroblocks in which the picture is divided, with an exception of peripheral macroblocks;
• calculating an average smoothness index for a preestablished number of last processed pictures;
• monitoring a number of macroblocks belonging to a same picture coded as intra macroblocks;
• identifying a change of a scene from a combination of the number of macroblocks belonging to the same picture coded as higher intra compared to a threshold number and/or of a variation of the smoothness index of two successive pictures compared to a top and a bottom threshold based on the average value of the smoothness index of the preestablished number of last processed pictures;
• and
• using identification of the change of the scene to control prediction computation of pictures using only a forward motion estimation for pictures preceding the change of the scene, and using only a backward motion estimation for pictures subsequent the change of the scene.

An MPEG encoder comprising:

• a motion estimation system for common fields of successive pictures, said motion estimation system comprising an algorithm of motion estimation responsive to identification of an intervened change of a scene, said algorithm for causing said motion estimation system to perform the steps of:
• defining and calculating a smoothness index of a motion field of each picture by analyzing motion vectors of all the macroblocks in which the picture is divided, with an exception of peripheral macroblocks,
• calculating an average smoothness index for a preestablished number of last processed pictures,
• monitoring a number of macroblocks belonging to a same picture coded as intra macroblocks,
• identifying a change of a scene from a combination of the number of macroblocks belonging to the same picture coded as higher intra compared to a threshold number and/or of a variation of the smoothness index of two successive pictures compared to a top and a bottom threshold based on the average value of the smoothness index of the preestablished number of last processed pictures, and
• using identification of the change of the scene to control prediction computation of pictures using only a forward motion estimation for pictures preceding the change of the scene, and using only a backward motion estimation for pictures subsequent the change of the scene.