線性內插法輔以非參數化極點同色異譜應用於多樣材質頻譜重建

Abstract

隨著數位影像的發展與演進，對於色彩複製的精準度要求日漸提升。然而，目前大多數硬體設備所採用的三維色彩數值，其色彩會受光源、物體特性、觀測者色匹配函數的影響，在影像處理的鏈結中轉換使用較不切實際。因此，物體頻譜反射率的量測與重建技術備受產業與學術界關注，若能採用不受觀測條件影響的物體頻譜反射率作為色彩描述的數值，實有助於色彩在各種觀測條件下的重現。 本研究提出以線性內插法為基礎的物體頻譜反射率重建演算法(INMSEP)，並輔以非參數化極點的同色異譜。以常見的三色式相機取得的三維色彩數值模擬高維度的頻譜資訊，將提出的方法應用於重建常見的色票材質和其他多樣材質樣本的頻譜反射率，與典型主成分分析法(PCA)、權重主成分分析法(wPCA)和利用模式化色域極點同色異譜的自然鄰點內插法(MMSEP+NNI)進行比較，並以均方根誤差、曲線配適係數及A光源及TL84光源下的色差評估重建結果。實驗結果表明，INMSEP方法在重建色票和壓克力顏料等人造材質樣本的頻譜反射率具有較高的準確性，然而，相對於基於PCA的方法，INMSEP方法在重建自然材質樣本的效果並沒有提升。綜合來看，INMSEP方法與wPCA方法在頻譜重建的實用性考量上都十分良好。With the development and evolution of digital images, the demand of accurate color reproduction has been increasing. However, most of the hardware equipment uses the three-dimensional color values, meaning that color appearance could be affected by the illuminant, object characteristics, and color matching function. In terms of the color conversion of image processing link, the usage of three-dimensional color values would be considered unrealistic. Therefore, the measurement and recovery of spectral reflectance of object has drawn great attention to the industry and the academic community. As the spectral reflectance of object is inactive to the viewing condition, it could be applied as the color description, resulting in the more efficient reproduction of the color under various viewing conditions. This paper presented a recovery of spectral reflectance method based on linear interpolation(INMSEP), supplemented by 8 nonparametric metameric spectra of extreme points. The three-dimensional color values obtained by the common trichromatic camera are used to simulate the spectral information of high dimension. We conducted experiments that selected the color chips material and compared the reconstructed results of various types of materials. The effectiveness of this suggested method was evaluated with a comparison of the classical PCA method, weighted PCA method, and natural neighbor interpolation method with model-based metameric spectra of extreme points (MMSEP+NNI). There is a color difference under illuminant A and TL84. The RMSE and the GFC values are used as the criteria of the recovered spectrum. The experimental results show that the INMSEP method has high accuracy in reconstructing the spectral reflectance of artificial materials such as color chips and acrylic pigments. However, compared with the PCA-based method, the INMSEP method does not have the better effect of reconstructing natural material samples. In summary, the INMSEP method and the wPCA method are very good in the practical consideration of spectrum reconstruction.