基於GPU邊緣導向的影像內插法

GPU-BASE EDGE-DIRECTED IMAGE INTERPOLATION

侯孝偉(Hsiao-Wei Hou)

指導教授 ： 鄭嘉慶

大同大學/電機資訊學院/通訊工程研究所/碩士(2009年)

摘要

影像的內插演算法可對影像進行縮放，在現今的數位顯示技術中這是一項非常重要的技術。一般的傳統內插法如最鄰近內插法、雙線性內插法以及雙立方內插法等，此類演算法的原理是建構在信號的連續性上，因此當遇到信號不連續時，也就是在邊緣部份時，往往會產生嚴重的失真，例如邊緣的格狀效應以及影像的模糊效應。
為了改善上述的失真，本文提出一個基於邊緣導向的影像內插法來進行影像內插。我們的方法主要是改進iNEDI演算法，在iNEDI中，圖像被分成邊緣區域與非邊緣區域兩部分，對於邊緣區域的像素，iNEDI使用圓形且可變的遮罩來求取內插係數，在非邊緣部份的像素，則是使用雙立方內插法來進行內插。我們的方法主要是改進iNEDI在非邊緣部份的內插，我們採用功能相當但運算教快速的二階多項式內插法來求取內插值。
另外我們也使用圖像處理單元來大幅減少運算時間，在執行上使用CUDA加速的Matlab平台實現。實驗結果顯示我們所提出的方法，在影像放大的品質與計算時間上都優於iNEDI演算法。

關鍵字

影像內插 ； 超解析 ； 邊緣偵測 ； GPU

並列摘要

Image interpolation algorithm for image scaling plays an important role in modern digital display technology. Traditional interpolation methods such as the bilinear interpolation and bicubic interpolation are based on the continuity of the signal. Therefore, when the signal is discontinued, e.g., for edge pixels, it tends to have serious distortions, such as blocking and blurring effects.
In order to solve these unpleasant distortions, In this thesis we propose an edge-directed interpolation method for image interpolation. Our method is an improved version to the iNEDI algorithm. iNEDI classifies an image into two categories of pixels, edge and non-edge pixels. For edge pixels, iNEDI computes the interpolated value using its four nearest neighbors with weighted coefficients estimated from circular masks with variable sizes. For non-edge pixels, it applies the bicubic interpolation to determine the interpolated value. Our proposed method focuses on the improvement of iNEDI in interpolating non-edge pixels. We proposed a second-order polynomial interpolation method whose effectiveness is comparable to that of bicubic method; yet it is more efficent than the bicubic method.
Furthermore, we use the graphics processing unit to speed up the computing by adopting the nVidia Cuda to accelerate the cumputation of Matlab. Experimental results show that our proposed method performs better than iNEDI in balancing both efficiency and effectiveness.

並列關鍵字

Edge-directed interpolation ； Super-resolution ； Image interpolation ； GPU

參考文獻

[1] A. Pratt, and K. William, Digital Image Processing, Wiley, N.Y., 1978.

[2] R. Keys, “Cubic convolution interpolation for digital image processing,” IEEE Trans. Signal Processing, vol. 29, no. 6 , pp. 1153 - 1160, Dec,1981.

[3] J. Allebach and P. W. Wong, “Edge-directed interpolation,” in Proc. IEEE Int. Conf. Image Processing, vol. 3, pp. 707–710,1996.

[4] Q. Wang and R. Ward, “A new edge-directed image expansion scheme,’ in IEEE Proc. of ICIP 2001, vol. 3, pp. 899-902, 2001.

[5] J. W. Hwang and H. S. Lee, “Adaptive image interpolation based on local gradient features,” IEEE Signal Processing Letters, vol, 11. no. 3, pp. 359 – 362, 2004.