In a movement compensation predictive encoder, a vector detector (19) is supplied with a moving picture signal and produces a movement vector signal representative of at least one movement vector for each region of each frame. A picture analyzer (16) serves as a variation detector for detecting a variation in each region to produce a variation signal when the variation exceeds a predetermined threshold in one of the regions. The variation signal is used in giving a predetermined value, such as zero, to the above-mentioned at least one movement vector for that one of the regions. In place of the variation detector, a scene change detector can likewise be used to produce a scene change signal for giving a predetermined value to the movement vectors for a frame for which a scene change is detected. The scene change detector is preferably operable on a summation for each frame of results of evaluation for movement compensated prediction errors rather than, as in prior art, for differences between picture elements of two consecutive frames.
A movement detector necessary for processing composite color television signals, which detects movement in image signals, wherein the detecting circuit is so constructed that no deficient motion is produced. A difference signal between the present frame and that preceding by two frames of the television signals is converted into a motion information signal and a movement detection signal is made by integrating the signal obtained by the conversion with respect to temporal axis.
A correlation surface is derived by phase correlating two pictures selectively displaced in the X and Y directions. The illustrative surface shows a large peak at zero displacement, corresponding to a stationary background, and a fairly large peak corresponding to a moving object, the X, Y position of the peak indicating the magnitude (pixels per field period) and direction of the motion vector. A set of motion vectors is thus determined and testing is then carried out, on a pixel by pixel basis or pixel block by pixel block basis, to determine which of the motion vectors gives the best match in deriving the second picture from the first. The motion vector thus assigned may be used in temporal interpolation of the pictures.
A redundancy reduction algorithm, such as hybrid DCT, is applied to a video signal and is subsequently encoded by pixel blocks, and motion vectors are generated for each image block. At least one global motion vector is then generated for the entire image, the block motion vectors are substracted from the global vector to obtain local motion vectors representing the shifts from the global vector, the global vector is encoded with fixed length, the local vectors are encoded with variable length, and the coded global vector and the coded local vectors are multiplexed into the coded video signal to be transmitted.
An MC predicting apparatus which ensures a high prediction accuracy against input image signal stream and avoids arithmetic operation of an evaluation function from becoming an overhead for an image encoding process is provided. Sequence numbers given to respective frames are discriminated, and a degree of interframe correlation is decided from a difference in sequence number of the preceding and current frames. A reference block generator adaptively controls, depending on the degree of interframe correlation, the reference range of MC prediction of the preceding frame stored in a frame memory. As a result, the number of times of arithmetic operation of an evaluation function for MC prediction can be reduced and a prediction accuracy can be improved.
In a high-efficiency encoder which performs motion-compensation prediction, an intra-field is set every n fields. The presence of a scene change is detected. When a scene change occurs, a reference picture of motion-compensation prediction is switched, or the field immediately after the scene change is set as an intra-field.
