This thesis is concerned with some basic problems of Sub-Band Coding (SBC) for image data compression. They are the quadrature mirror filter (QMF) design, filtering method for the boundary region of a limited image, classified vector quantization (CVQ) and motion estimation schemes of the sub-banded image data.
A new design technique for linear phase QMF pair is proposed for the frequency responses of sharp and symmetrical transition characteristics. It has been shown that the filters has the capabilities of perfect reconstruction and simple implementation. The simulation results show that the new filter pair performs as good as longer conventional QMF pair in the presence of quantization errors.
In order to reduce the boundary effect resulted from filtering of the boundary region of a limited image, a now filtering technique called loop convolution is presented. This method satisfies the 'alias-free' and 'error-free' requirement in the reconstructed image. It has been shown that the loop convolution has better performance than the conventional circular convolutions in the presence of quantization errors.
An edge-based classification based on two sub-band pels is proposed for CVQ of sub-banded images. It is found to be effective in overcoming the edge degradation in the coded images at the low rates. In Hierarchical SBC(H-SBC) for sequential image coding, the effect of hierarchical block motion estimation using pass-band pyramid is investigated. It is found that the new H-SBC scheme can not only exactly estimate the motion but also prevent the cumulation of quantization errors in the pyramidal image reconstruction.