| Summary: | 碩士 === 國立高雄大學 === 電機工程學系碩博士班 === 105 === Evolvable hardware (EHW), which is a combination of reconfigurable hardware and evolutionary algorithm, is an emerging research topic.Recent studies show that the integration of fuzzy theory and EHW demonstrates effectiveness on digital image filtering.However, in these studies fuzzy rules and membership functions are defined heuristically. Their performance on filtering a variety of types of image noise is limited.
%%
In this study, similarity and divergence of image pixels are analyzed and used for clustering.Each cluster is defined as a fuzzy classification rule.The belongingness of pixels to a cluster describes a model of fuzzy membership function in terms of similarity and divergence.
%%
A clustering-based incremental algorithm is developed for generating fuzzy rules and membership functions from a given set of image pixels.With each fuzzy classification rule, an EHW-based image filter is learned and used for filtering the pixels classified by the fuzzy rule.Because fuzzy rules are learned from image pixels, not defined statically, our proposed method can have better performance on processing noisy pixels with the correct circuits.
%%
In this study, the performance of our proposed method is compared with other ones.The experimental results show that our proposed method outperforms traditional methods on image filtering.\\
|
|---|
