| 摘要: | 數位相機的發展使每個人都可以隨時捕捉生活中的影像，但其所見不為所得與同色異譜的問題在色彩複製的領域中，是一直持續研究的議題，藉由多頻譜影像技術的產生，對於影像色彩的記錄方式由原本的RGB影像改為以可見光頻譜反射率能量算出ZYZ後轉換到Lab的方式所計算出的多頻譜影像，解決了同色異譜的問題，但卻造成龐大資料量而不易保存。
本研究以標準的JPEG、JPEG2000的壓縮演算法以及相關多頻譜的文獻探討為發想，針對多頻譜影像做一維與二維兩種維度的切割方式，使用離散餘旋轉換與小波轉換做空間資訊與頻率資訊間的轉換，並以閥值與量化兩種方式做資料壓縮，共十四種組合做為本研究所使用的壓縮方法。以Mini ColorChecker® 24色導表對此十四種組合做測試比較後，挑選色差值較佳的六種組合壓縮方法，對其他不同類型影像做驗證，最後歸納出三種平均表現較佳的組合壓縮方式。
本研究以色差做為主要的影像品質標準，提供多頻譜影像壓縮倍率與色差之間的品質關係，在研究中搭配PSNR與MSE兩種影像品質計算方式，其實驗結果為使用一維壓縮方法在平均色差小於1的影像，壓縮率可達8~10倍，以二維壓縮方法可達30~50倍，而PSNR均為35以上，MSE為3.0E-4以下，可提供與其他多頻譜壓縮文獻的比較依據。
The problem of metamerism is commonly seen in color reproduction by regular tri-chromatic (RGB) camera, where color-matchings can appear differently under different conditions. This problem can be solved by multi-spectral image technology by processing object’s spectral reflectance. Multi-spectral image uses higher color fidelity than RGB image so it takes up much more memory space than regular tri-chromatic image.
This study adopts JPEG and JPEG2000 compression algorithms to produce fourteen kinds of compress methods in three kinds of combination: selecting one or two dimensional slicing, using DCT or DWT transformations, selecting threshold or quantification methods.
At first stage, a GretagMacbeth Mini ColorChecker was used as the index to pick the better six compression methods by having smaller color difference values before and after compression. Secondly, these six compression methods were compared by using six multi-spectral images captured from a spectral camera and were reduced to three kinds of compression methods having lower averaged color difference.
In this study, the color difference values were used as a primary index to determine the quality of a compression method. Meanwhile, PSNR and MSE were also calculated for references. The results indicate that a compression rate of 8 to 10 can be achieved for color difference value less than 1 when using one dimensional compression method. A compression rate of 30 to 50 can be achieved when using two dimensional compression method. All the PSNRs are higher than 35 and MSEs are below 3.0E-4 below for the final 3 compression methods. |
