採用優化的動態像素位移演算法達到解析度提升

Optimized E-shifting Algorithm for Resolution Enhancement

簡彗羽(Hui-Yu Chien)

指導教授 ： 黃乙白 ； 田仲豪

國立交通大學/電機學院/光電工程系所/碩士(2019年)

摘要

近眼顯示器提供了全新的觀賞模式給使用者。傳統近眼顯示器採用雙眼視差，使左右眼分別接收不同訊號，在腦中結合並產生立體影像。然而這種方法會因為單眼對焦和雙眼聚焦之間的差異，而使人們感到頭暈。在2013年，光場近眼顯示器被提出其可以克服上述的問題。有別於傳統的近眼顯示器，在光場近眼顯示器中，觀賞者可藉由回追光線，看見重建的3D影像，其中單眼對焦和雙眼聚焦的距離相同。也因此，光場近眼顯示器克服了單眼對焦和雙眼聚焦之間的差異。然而，光場近眼顯示器的解析度卻低很多。在之前的研究中，運用了時間多工的方法。在硬體方面：添加了扭轉型液晶做偏振切換和雙折射率材料使影像位移；在軟體方面：提出了配合硬體的動態像素位移演算法。結合硬體與軟體，讓影像在人眼的視覺暫留機制下，達到解析度提升的效果。實驗結果顯示相比於傳統的光場顯示器，其光學解析度有效地提升了1.4倍，並且可以減緩因為透鏡放大而導致的柵格化影像。
在本研究中，提出了運用更多元的影像資料庫去優化了先前的動態像素位移演算法。在提出的動態像素位移演算法中，為了實現傳統取樣時的內插法，結合了「提升倍率的方法」、「提升的倍率」，以及「取樣方法」去做分析。藉由拆成三個步驟，一一優化，達到傳統內插法的目的。在分析過程中，有五種提升倍率的方法。其中Edge directed unsharp masking sharpening (EDUMS)方法達到較好的分析結果，因此採用此法作為本動態像素位移演算法中的提升倍率的方法。再者，本研究中，提升的倍率為1.3倍至2.5倍。最後，有兩種取樣方法，其為配合各種提升倍率的應用結果。
此外，為了更完善地發揮各提升倍率的優點，我將影像資料庫依照空間頻率成分分類為四種情況。並對四個分類分別用不同提升倍率的動態像素位移演算法。最終，獲得兩種各自表現於高空間頻率影像以及低空間頻率影像最好的方式。不僅能使高空間頻率影像的邊緣更清晰，也能維持影像的解析度。相較於Full HD面板的原生解析度，解析度仍維持1.78倍的提升；而在光場近眼顯示器，卻有1.59倍的提升。

關鍵字

解析度提升 ； 動態像素位移 ； 動態像素位移取樣演算法

並列摘要

Near-eye displays (NED) provide whole new viewing experiences for users. Most traditional NED create 3D images by using binocular parallax method. The right eye and the left eye receives different images, and the human brain will generate 3D images by merging the images. However, this method causes traditional NED to suffer from accommodation-convergence conflict (A.C. conflict), and it will cause visual fatigue to users. To solve the issue, light field near-eye displays (LFNED) was proposed in 2013. Unlike traditional NED, in the LFNED, viewers can see a reconstructed 3D image by tracing rays back through the image plane with the same accommodation distance and the convergence distance. Therefore, LFNED eliminate A.C. conflict. However, LFNED suffer from the low spatial resolution. In the recent work, a time-multiplexed method was proposed. The hardware includes a birefringent plate and a twisted nematic liquid crystal plate (TN plate) which can achieve image shifting. The software includes an e-shifting sampling algorithm. With the hardware and the software, images will be enhanced by the visual persistence. The results show that the resolution was enhanced for 1.4 times. Moreover, it can reduce screen-door effect and rasterized effect to get smoother image.
In this research, using a diverse database to optimize the e-shifting sampling algorithm is proposed. To achieve the traditional interpolation in the proposed e-shifting sampling algorithm, I divide it into three steps: the upscaling methods, the upscaling ratios, and the downsampling methods. The traditional interpolation will be replaced with the optimization of each step. In the analyses of the optimization, there are five upscaling methods. The edge directed unsharp masking sharpening method is chosen as the upscaling method in the proposed e-shifting sampling algorithm, because of its higher PSNR value and higher spatial frequency at the same contrast. Moreover, the upscaling ratios in this research are from 1.3 to 2.5 times. At last, there are two downsampling methods which are corresponding to the upscaling ratios.
In addition, to develop the strength of each upscaling ratio, I categorize the image database into four cases by the spatial frequency. And practicing the e-shifting sampling algorithm with the different upscaling ratios to four cases. Finally, the best e-shifting sampling algorithm for the high-spatial frequency images and the best e-shifting sampling algorithm for the low-spatial frequency images are presented. The edges of the high-spatial frequency images are sharper than the ones by the previous e-shifting sampling algorithm, the resolution is still high at the same time. Comparing to the native resolution of the Full HD panels, the resolution of the proposed e-shifting method is enhanced for 1.78 times of the images. Also, the resolution of the proposed e-shifting method is enhanced for 1.59 times in the light filed near-eye displays.

並列關鍵字

Resolution enhancement ； Wobulation ； E-shifting ； E-shifting sampling algorithm

參考文獻

1. Chavan, Sagar R. "Augmented reality vs. virtual reality: differences and similarities." (2014).

2. Wheatstone, Charles. "XVIII. Contributions to the physiology of vision.—Part the first. on some remarkable, and hitherto unobserved, phenomena of binocular vision." Philosophical transactions of the Royal Society of London 128 (1838): 371-394.

3. Aukstakalnis, Steve. Practical Augmented Reality: A guide to the technologies, applications, and human factors for AR and VR. Addison-Wesley Professional, 2016.

4. Ukai, Kazuhiko, and Peter A. Howarth. "Visual fatigue caused by viewing stereoscopic motion images: Background, theories, and observations." Displays 29.2 (2008): 106-116.

5. Hoffman, David M., et al. "Vergence–accommodation conflicts hinder visual performance and cause visual fatigue." Journal of vision 8.3 (2008): 33-33.