A Study on Particle Swarm Optimization based Resistant Fractal Image Compression using Least Trimmed Squares
| Autor: | I-lin Wu, 吳易霖 |
|---|---|
| Rok vydání: | 2009 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 97 Fractal image compression (FIC) is a lossy coding scheme. It possesses the advantages of high quality of retrieved image, zoom invariant, and high compression ratio. It is used for many applications recently in the filed of image reconstruction, watermark, medical image, feature recognition, and so on. However, if a corrupted image is encoded by FIC, the quality of retrieved image will be poor. Self-similarity and partitioned iterated function system are the underlying idea of FIC. Practically, we find the similarity between range blocks and domain blocks in the encoding process of FIC. The scheme of least squares is used to calculate contrast and brightness between a range block and a domain block in the conventional FIC. The least squares estimator is the best linear unbiased estimator under assumptions that the random variable is zero-mean, constant variance, and uncorrected random variable. As is well known in regression theory that linear regressor is sensitive to outliers. That’s reason why the quality of retrieved image will be poor. Robust regression is usually used against noises in regression theory. Least trimmed squares (LTS) method with resistance from the robust regression is proposed in this thesis. It is embedded into the encoding procedure of the FIC. Recursive weighted least squares with simple architecture and fast convergence is used in this thesis. To effectively improve the encoding speed, particle swarm optimization (PSO) is utilized to reduce the search space. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|