Nonlinear algorithms for image resolution enhancement and image compression.

摘要

The main goal of our research is to study and investigate the high potential of the nonlinear algorithms alternative to classical linear algorithms in certain image related applications such as still image compression and image resolution enhancement (still image interpolation, superresolution from low resolution videos). We propose several novel nonlinear algorithms and demonstrate their promising advantages.; More specifically, in the area of still image interpolation, we present a novel fast adaptive nonlinear algorithm to enhance the resolution of an image by an arbitrary factor. The scheme significantly outperforms the classical linear interpolation schemes such as bilinear and bicubic. We also compare our scheme with other adaptive interpolation schemes. The proposed scheme achieves comparable visual quality with lower computational complexity.; In the area of superresolution from low resolution videos, we present an efficient multi-stage block matching motion estimation algorithm to determine the correspondence among low resolution frames. The information obtained from motion estimation is used to reconstruct the high resolution image under the POCS superresolution framework. Furthermore, we propose new edge-related constraints inside the POCS reconstruction to suppress certain artifacts. Simulation results are provided to compare the performance of our motion estimation and the hierarchical block matching (HBM) motion estimation technique. We also show the reduction of artifacts by utilizing our edge-related constraints in the POCS reconstruction.; In the area of still image compression, we present a novel algorithm to compress an image under the nonlinear non-expansive image pyramid framework. Our performance of lossless compression is comparable or only slightly worse than the best lossless compression algorithms such as LOCO-I and CALIC. However, it has the additional feature of multiresolution format which is not present in CALIC and LOCO-I. In general, the algorithm has lower numerical performance in lossy compression compared to SPIHT, while it has superior performance on mixed text/graphics images over SPIHT and JPEG.