There are two important steps in high quality coding of video image data. The one is the estimation of objects or block motion based on BMA(Block Matching Algorithm), which is to compensate for the motion of objects in successive images, and the other is the predictor selection method, which selects a better predictor among several predictors, depending on data and motion.
In this thesis, a new BMA which estimates the motion with the menu vector set chosen by using mean and variance as block features is proposed. Its performance is compared with that of the other existing BMA's. In comparison of prediction accuracy, the entropy of prediction error and SNR(Signal-to-Noise Ratio) are used as performance measures.
Using the black features which are not considered in the other methods, the proposed method needs additional hardware, but it can be also applied to other image coding techniques, for example, codebook search in vector quantization of image data.
The computer simulation results show that the performance of the proposed BMA is better than that of the other existing ones, and the number of search points of the proposed is less than that of the others for similar performance. In addition, the fewer number of sequential steps in the proposed method makes real time processing more advantageous.
A predictor selection method which combines the existing predictor selection methods by using average concept is also proposed for effective selection. This can reduce the probability of wrong selections by the existing methods.
By using the proposed method, of SNR gain of 1 - 2 dB is achieved, and an entropy reduction of 0.1 ~ 0.2 bit/pel can be achieved in comparison to other methods. In the hardware complexity, the proposed method is a little more complex than the other existing ones.