誤差散播式半調色影像之紋路輪廓消除、影像強化、邊緣偵測及壓縮

Abstract

誤差散播式(error diffused)半調色(halftone)是一種影像編碼技術，可 以將灰階影像轉換成一種特殊的黑白二色影像(簡稱為ED影像)，其視覺上 仍保有原灰階影像之明暗漸近感。但是在ED影像中，平滑區域常會出現不 自然的紋路輪廓，而且細部線條也往往不夠清楚。此外灰階影像中若含有 突波雜訊，則會危害到所產生ED影像之品質，因此，在本論文中，我們提 出三個方法來改進ED影像的品質，它們的功能分別是：(1) 消除ED影像中 ，平滑區域常出現的不自然紋路輪廓，(2) 除了消除不自然紋路輪廓外， 同時再加入邊緣強化的功能，(3) 濾除灰階影像中之突波(impulse)雜訊 。最後，除了上述三種ED影像品質改善法外，我們也提出一個基於邊緣偵 測及邊緣保存觀念的ED影像壓縮方法。現有之方法需先將ED影像轉回灰階 影像再做壓縮，我們所提的技術則不受此限制，而可以直接的應用在ED影 像，而且實驗顯示這個快速的方法可以達到即時的運作。 Halftone techniques deal with the issue that transforms gray- level images into bilevel ones which visually preserve the gray aspect and image detail of the original gray-level images. Among the reported halftone techniques, error diffusion (ED) is a frequently used approach. However, the bilevel images generated by the ED tend to have false-texture contours in smooth areas, and edge enhancement is also needed to improve the quality of the halftone images. Besides, images are often corrupted by impulse noise due to a noisy sensor (or communication channel), therefore, an impulse noise removal procedure is usually needed to remove the impulse noise from the input gray-level image before the halftone process is applied. An important topic included in this study is about data compression. Although halftone images only need 1 bpp (bit per pixel) storage space, large images will still induce noticeable transmission delay. Consequently, a fast halftone image compression technique is expected. In this dissertation, we first propose two improved ED algorithms and a minimum-maximum exclusive mean (MMEM) filter for the applications of digital image halftone, followed by a compression technique directly applied to error diffused images. The first ED algorithm uses a local filter to smooth the false- texture contours, and the second algorithm not only smoothes the false-texture contours but also sharpens the images by combining Eschbach's edge-enhanced algorithm with the proposed artifacts reduction method. Experiments show that the results are satisfactory. As for the proposed impulse noise removal filter, the MMEM in a window is used as the estimation of the true value to restore the corrupted image. Experimental results show that even if the impulse noise is heavy (70%), the proposed filter can still work properly and the restored image is acceptable. Finally, in the compression of error diffused image, each input (ED) image is divided into non-overlapping blocks of size , and each highly-consistent block is encoded by its average illumination, and each lowly-consistent block (edge block) is transmitted directly using original bitmap. (Here, the edge block is detected using a specially designed edge detection technique for ED images.) The proposed approach is directly applied to the error diffused images without any inverse halftoning technique. The complexity of the proposed compression technique is low, and this fast method can be used in real-time application.