Process and device for global motion estimation in a sequence of images and a computer program product therefor

A system for performing an estimate of global motion based on a sequence of subsequent video images, such subsequent video images being represented by digital signals arranged in frames, comprising:

• a processor including:
• means for choosing amongst a series of vectors originating from linear combinations of motion vectors resulting from estimates of previous frames and/or constant vectors, at least one winning vector, considered as the best match for estimating the motion occurring between two frames, including:
• means for performing a virtual overlay of two frames to be compared mutually offset both horizontally and vertically by amounts identified by the motion vector subjected to testing;
• means for applying a matching mask of pixels to be subjected to testing;
• means for calculating, for all pixels selected, a cost function;
• and
• means for identifying the vector with the lowest cost function value as the best vector for the purposes of estimation.

A computer program product directly loadable into the internal memory of a digital computer and including software code portions adapted for implementing, when said computer program product is run on a digital computer, a process for performing an estimate of global motion based on a sequence of subsequent video images or the like, such subsequent video images being represented by digital signals arranged in frames, and, for each estimate of a frame with respect to another, comprising:

• choosing, from amongst a series of vectors originating from linear combinations of motion vectors resulting from estimates of previous frames and/or constant vectors, at least one winning vector, considered as the best match for estimating the motion occurring between two frames, including:
• performing a virtual overlay of the two frames to be compared mutually offset both horizontally and vertically by amounts identified by the motion vector subjected to testing;
• applying a matching mask of pixels to be subjected to testing;
• and
• calculating, for all pixels selected via the grid, a cost function to determine the effectiveness of the predictor, identifying the vector with the lowest cost function value as the best vector for the purposes of estimation.