Spatial-spectral properties of discrete wavelet transform for image processing using Markov Random Field

The spatial-spectral properties of Discrete Wavelet Transform (DWT) in conjunction with Markov Random Field (MRF) are studied in this dissertation. There are two main areas of the research: the investigation of X-ray Wavelet Transform (XWT) with the directional MRF, and the investigation of DWT/Redundant DWT (RDWT) with non-directional Multiresolution MRF (MMRF). Both of them emphasize the research on the spatial-spectral properties of DWT with MRF models.; XWT has the ability to reveal the sudden changes along its transform direction. Directional MRF model emphasizes the textures in certain directions during the segmentation procedure. The properties of XWT and directional MRF in view of Digital Signal Processing (DSP) applications are investigated, and algorithms that use these properties efficiently in the directional clutter removal applications are presented in this dissertation.; Based on the features of MMRF and DWT, the DWT/RDWT based MMRF segmentation algorithm is introduced to process non-directional signal. In this algorithm, DWT is used to generate the pyramidal structured multiresolutional image where each LL subband image is the input for MMRF segmentation. The bandpass property of DWT in the spectral domain is studied in this dissertation and it is proved that the proposed algorithm is very suitable for segmentation of images with regard to the global features without detailed local texture. Based on this scheme, algorithms are designed in areas of tumor detection for breast cancer, Region-of-Interest (ROI) based image compression for remote sensing images and object-oriented video segmentation. The results prove the effectiveness of the designed algorithms.