Method for block matching motion estimation in digital video sequences

#This invention relates to methods for block matching motion estimation in digital video sequences.

#Research on digital video communication has received continuously increasing interest over the last decade. Some of this work converged into robust standards successfully set on the market, (MPEG1 and MPEG2; H.261 and H.263), and a new very powerful multimedia standard (MPEG4), that is mainly based on the idea of second generation video coding. However higher bandwidth availability in the communication channels, e-commerce applications over the web and the video e-mail will certainly open new horizons for digital video.

#Even though it is possible today to execute an MPEG playback in real time over most of the personal computers available, the situation worsens considerably when we consider the video encoding. It is computationally so demanding that it becomes simply impossible to find encoders able to work in real time, with a decent quality, without specific added hardware.

#Other objects, uses and advantages of this invention will be apparent from the reading of this description which proceeds with reference to the accompanying drawings forming part thereof and wherein:

#FIG. 1 shows a general motion compensated coding and decoding structure,

#FIG. 2 shows a flowchart of the SSA (Statistical Search Algorithm) according to the invention,

#The general structure of a motion-compensated codec is shown in FIG. 1. For the moving areas of an image, there are roughly three processing stages at the encoder: motion parameter estimation, moving pels prediction (motion compensation), and prediction-errors/original-pels compression. Only two stages are needed at the decoder: prediction-errors/original-pels decompression, and motion compensation.

#In these processes, the motion compensators at the encoder and decoder sides are similar. They often consist of two operations: access to the previous-frame image data according to the estimated displacement vectors, and construct the predicted pels by passing the previous frame image through the prediction filter. The data compression and decompression processes are inverse operations of each other. The most popular image data compression procedure uses discrete transforms, coefficient selection quantization, and variable word length coding.

#FIG. 1 shows a general motion compensated coding structure with an encoder part to the left and a decoder part to the right of the dotted line. This structure is in principle known from the prior art. In the following description most of the emphasis will be put on the encoder part more precisely on the displacement estimator 2.