Methods and apparatus for performing dual prime motion estimation on video fields or frames of a video signal. A 16.times.16 motion estimator generates a same-parity match vector M.sub.-- SP and an opposite-parity match vector M.sub.-- OP for a current macroblock of a current field using a reference macroblock from each of a previous same-parity and opposite-parity field. A dual prime motion estimator receives the M.sub.-- SP and M.sub.-- OP match vectors for the current macroblock, and generates a base motion vector MV and a delta motion vector DMV for the current macroblock. In a first embodiment, the dual prime motion estimator generates MV and DMV by setting MV equal to M.sub.-- SP and then selecting DMV such that the opposite-parity vector OPV to be computed by a decoder is as close as possible to M.sub.-- OP. In a second embodiment, OPV is set equal to M.sub.-- OP and MV and DMV are selected such that MV is as close as possible to the same-parity match vector M.sub.-- SP. In a third embodiment, the dual prime motion estimator uses error measurements between the current macroblock and the same-parity and opposite-parity macroblocks to determine whether the techniques of the first embodiment or the second embodiment should be applied to the current macroblock.
In order to prevent accumulation of rounding errors caused by the integer-based bilinear interpolation used in the motion compensation process of image coding and decoding methods, two types of P frames are used for unidirectional motion compensation prediction: P+ frames, which round the results of bilinear interpolation with real number operations to the nearest integer and round half integer values (0.5 added to an integer) away from zero; and P- frames, which differ from P+ frames in that the above mentioned half integer values are rounded towards zero. Utilizing both of these P+ and P- frames enables canceling of rounding errors and prevents the accumulation of rounding errors.
A method for encoding mode signals of a target block of M.times.N pixels having a binary value `0` or `255` is provided. If the target block is determined as neither "all.sub.-- 0" nor "all.sub.-- 255", the target block is divided into a top field and a bottom field, wherein the top field contains every odd row of the target block to have M/2.times.N pixels and the bottom field contains every even row of the target block to have M/2.times.N pixels. After encoding the top field, the method encodes the bottom field based on the top field to generate modified bottom field-coded data and a modified bottom mode.
A control circuit specifies a first search area by using a motion vector found in units of macroblocks between a plurality of frames located previous to first frame, specifies a second search area by using a simplified motion vector indicating the motion of an image between the first frame, the second frame, and a frame located in time between the first frame and the second frame, and designates a search area by switching the first search area and the second search area suitably.
An image predictive decoding method in which when compressively coded image data including information about a plurality of motion compensation modes, and a plurality of motion vectors corresponding to the plurality of motion compensation modes is decoded, for a target region to be decoded, target predictive region data calculated based on the motion vector of the target region, and adjacent predictive region data calculated based on the motion vector of at least one adjacent region to the target region, are weightedly averaged to produce optimal predictive region data to the target region, wherein the optimal predictive region data is produced using only the motion vector of the adjacent region having the same motion compensation mode as that of the target region. Thereby, when overlapped motion compensation is performed to an interlaced image, a highly efficient predictive signal is generated, and a coding efficiency is improved.
A method for motion coding an uncompressed digital video data stream such as an MPEG-2 digital video data stream. The method includes the steps of comparing pixels of a first pixel array in a picture currently being coded with pixels of a plurality of second pixel arrays in at least one reference picture and concurrently performing motion estimation for each of a plurality of different prediction modes in order to determine which of the prediction modes is an optimum prediction mode determining which of the second pixel arrays constitutes a best match with respect to the first pixel array for the optimum prediction mode, and, generating a motion vector for the first pixel array in response to the determining step. The method is implemented in a device such as a motion estimation search system of a digital video encoder. In one embodiment, the method and device are capable of concurrently determining performing motion estimation in each of the six different possible prediction modes specified by the MPEG-2 standard.
