臺灣大學: 資訊網路與多媒體研究所傅楸善余浚弘Yu, Jun-HongJun-HongYu2013-03-222018-07-052013-03-222018-07-052011http://ntur.lib.ntu.edu.tw//handle/246246/251426大部分的數位相機為了降低製造成本，使用了單一個CCD/CMOS 陣列，並搭配一個色彩濾波陣列(CFA) 來取得數位影像。此論文提出了一個嶄新的高品質放大演算法。首先根據此適應性異質性投影遮罩及植基於索貝爾及內插的遮罩，我們可以更精準地抽取馬賽克影像上的邊資訊，像是變化方向及梯度資訊等。我們可以根據這些資訊先將影像的綠色部分放大。接下來再根據色差平面結合內插的方法可以將影像的紅色和藍色部分放大。我們用了二十四張常被使用的測試影像來進行實驗。實驗結果顯示，以PSNR 及CPSNR 這兩個量度標準來看，我們的演算法所產出之放大影像品質較幾個先前發表的放大演算法要好，同時色彩瑕疵也更少。Mosaic images are captured by a single charge-coupled device/complementary metal-oxide-semiconductor (CCD/CMOS) sensor with the Bayer color filter array. In this paper, a new quality-effective zooming algorithm for mosaic images is presented. First, based on adaptive heterogeneity projection masks and the Sobel- and lumi-nance estimation-based (SL-Based) masks, more accurate gradient information can be extracted from the mosaic image directly. According to the extracted more accurate gradient information, the mosaic green (G) channel is first zoomed. In order to reduce color artifacts, instead of directly moving the original red (R) value to its right position and the blue (B) value to its lower position in the previous approach, the color difference interpolation is utilized to expand the G-R and G-B color difference values. Finally, the zoomed mosaic R and B channels can be constructed by using the zoomed G channel and the two expanded color difference values; afterward, the zoomed mosaic image is obtained. Based on twenty-four popular test mosaic images, experimental results are demonstrated to show that the proposed zooming algorithm has more than 1.79 dB quality improvement when compared with two previous zooming algorithms, one by Battiato et al. and the other by Lukac et al.1608892 bytesapplication/pdfen-US色彩濾波陣列(CFA)適應性異質性投影色差平面數位相機邊資訊梯度資訊馬賽克影像索貝爾算子放大Color differenceColor filter arrayGradient informationMosaic im- agesPSNRSobel operatorZooming algorithmhttp://ntur.lib.ntu.edu.tw/bitstream/246246/251426/1/ntu-100-R96944026-1.pdf