| Summary: | 碩士 === 國立成功大學 === 電機工程學系 === 88 === Wavelet-based coding, such as the new still image compression standard JPEG2000, plays an important role in image data compression. However, wavelet-based coding produces noticeable image degradation, the so called “ringing artifacts”, at the vicinities of edges in the reconstructed images, especially when the images are highly compressed. Ringing artifacts result in artificial contours and so degrade the quality of a decoded image. To improve the visual quality of decoded images and make them perceptually acceptable, post-processing for de-ringing is often used.
In this Thesis, an adaptive directional morphological filter (ADMF) to reduce the ringing artifacts is presented. It uses in total eight structural elements (SEs), which are classified into two categories, to remove ringing artifacts and preserve image features meanwhile. Each category contains four SEs for four possible directions. A texture detector is used to detect possible image features, such as thin lines and edges, and decide the category of SEs used. By texture detection, image features and ringing artifacts are filtered by different SEs so that the edges are well preserved, i.e., not blurred. The test images include various natural and synthetic images. As compared with the previous proposed methods, ADMF is rather efficient and effective. Simulation results show that ADMF is capable of significantly reducing the ringing artifacts.
|
|---|
