Abstract: Motion correlation is a way of analyzing movement in image sequences such as television signals. The estimation of movement in television pictures is essential to enable the use of motion compensated processing techniques. These techniques yield improved quality video processing and moving image compression. Motion correlation would be used as part of a video motion estimation system. This invention describes a number of enhancements to, and a simplfied, efficient implementation of, the basic motion correlation algorithm. It is shown how to produce motion correlation surfaces co-timed with output, rather than input, pictures. It is also shown how to average power spectra to obtain improved noise performance, and hence improved accuracy, in motion correlation. Finally it is shown that separate correlation surfaces can be produced relating to `future` and `past` pictures.

Abstract: At the transmitter a coder includes a motion vector generator providing vectors (MV) describing the movement of individual blocks of pixels. The video signal is 4:1 compressed in bandwidth by pre-filters and a sub-sampling unit to produce a signal (SSH) from which a high-definition image can be re-constructed in the coder and in the decoder. The sampling lattice is shifted in accordance with the motion vectors (MV) which are digitally transmitted along with the compressed bandwidth analogue signal, to enable the samples to be correctly located in the reconstructed image. Poorly correlated moving areas are handled by pure spatial filtering (pre-filter and sub-sampling unit) with reconstruction by spatial interpolation. The two reconstructed signals (RVH and RVL) at the coder are compared in a mode selector with the input video and a switch is set to transmit whichever of the compressed signals (SSH, SSL) gives the best match.