| 摘要: | 本論文介紹了數位全像儲存系統(digital holographic data storage system, DHDSS)的原理並提出了對位方法及辨識方法來改善數位全像儲存系統中的資料通道。在論文中，提出兩種對位方法並且比較其性能，一種是灰階權重法(gray level weighting method, GLWM)，另一種為相似結構定位法(alignment method based on structure similarity (SSIM), AMBS)。此外，辨識方法也是透過相似結構法(SSIM)法展而成的，稱為相似結構辨識法(recognition method based on SSIM, RMBS)。文中，我們會利用三個實驗來驗證這些方法的有效性。
GLWM是我們第一個提出用來改良數位全像儲存系統的對位方法。他利用資料通道中的灰階特性來準確的找出棋盤格影像上的定位點。但是，在GLWM中的邊界識別利用二值化方法完成可能會影響定位點的位置。因此AMBS利用影像品質評價的方法–SSIM來改正了GLWM的邊界識別方法來得到更準確的定位方法。而DHDSS也因為更準確的定位，誤碼率(BER)下降了1.5dB，當資料讀出時能有較大的位移寬容值。此外，當DHDSS利用低品質的光學元件來降低系統成本時，透過定位方法依然可讓系統具有令人滿意的BER。
RMBS利用一個影像資料庫儲存用來辨識的無噪聲影像，並像AMBS一樣，利用SSIM來提升辨識方法的性能，並嘗試去取代傳統的閥值法(二值化)。可惜的是，無噪聲的影像無法像一個相對映的參考影像來明確地辨識出一個具有嚴重雜訊的目標影像並且還原其原始資料。因此，在位移實驗中，RMBS的表現還需要更進一步的改進。儘管如此，RMBS在某些位移情況下仍優於閾值。
結合AMBS及RMBS這兩種方法可以削減DHDSS的系統成本，同時確保無錯誤的發生。作者希望這兩種方法的結合會使得DHDSS成為一個成熟的商業產品。
This thesis introduced the principle of the digital holographic data storage system (DHDSS), and proposed the alignment method and recognition method for improving the data channel in the DHDSS. There are two alignment methods proposed and compared in this thesis, one is gray level weighting method, GLWM, and another one is alignment method based on structure similarity (SSIM), AMBS. Also, the recognition method is developed by using SSIM, called recognition method based on SSIM (RMBS). Experiments are performed on three cases that demonstrate the effectiveness of these methods.
GLWM is the first method that is proposed herein to improve the alignment in the DHDSS. It uses a gray-scale, which is a characteristic of the data channel in the DHDSS, to locate accurately the fiducial point in a checkerboard image. However, an issue associated with boundary identification affects the location of the fiducial point. AMBS corrects a fault in the boundary identification, making it a more accurate alignment method to yield the accurate fiducial points in the data pages via an image quality assessment, SSIM. Since AMBS provides accurate alignment, the DHDSS has a 1.5 dB decrease in bit error rate (BER) and a greater tolerance for shift during the data readout. Also, inexpensive optical elements can be used in the DHDSS to reduce system cost while maintaining a sufficient BER.
RMBS is like AMBS in that its performance is improved by SSIM. RMBS uses an image database to store several noise-free images that are utilized in the recognition procedure. RMBS attempts to replace the thresholding method in the data channel by the effective SSIM. Unfortunately, the performance of RMBS in retrieving data from the data pages in Shifting case should be further improving because the noise-free images in the database do not unequivocally recognize the target image with serious distortion as a corresponding reference image. Nevertheless, RMBS still outperforms the thresholding method in some shifting cases.
Combining AMBS and RMBS can cut the system costs of the DHDSS, while keeping it error-free. The authors hope that the combination of the two methods will grow DHDSS into a finished commercial product. |
