In this thesis, novel compression methods for 2-D and 3-D medical images are proposed. The proposed methods provide better performance than the existing compression methods based on the wavelet transform.
The existing 3-D medical image compression techniques are less effective in case that the inter-slice distance is not narrow enough. In other words, their performance depends upon the inter-slice distance of medical images. In order to overcome this dependency problem, an adaptive intra/inter mode selection method is proposed. In this method, adaptive mode selection is performed in the wavelet domain rather than in the spatial domain to reduce unwanted blocking artifacts. Consequently, the proposed method provides good performance regardless of the inter-slice distance.
On the other hand, wavelet transform coding methods still show room for improvement. Because wavelet transform is independent of the characteristics of signals, if we find an adaptive transform according to either unknown or spatio-varying (or time-varying) characteristics of signals, it will improves the compression efficiency. The wavelet packet transform is regarded as one of such transforms. As a generalized version of the wavelet transform, the wavelet packet transform makes signals be concentrated into fewer coefficients than the wavelet transform does. However, its decomposed result is not a dyadic multiresolution structure like the wavelet transform. Therefore, zerotree coding, which provides an excellent performance, cannot be applied to the wavelet packet transform domain, since it utilizes the dyadic multiresolution structure. In this thesis, we propose a coefficient rearranging method, which enables zerotree coding to be directly applied to the wavelet packet transform domain. It provides performance improvement compared with the existing zerotree coding methods that are performed in the wavelet transform domain.
Finally, we combine the above two methods into a 3-D image compress...