應用二維經驗模態分解消除水域影像日光返照

Abstract

水域的航空遙測影像容易因為水面對太陽輻射的鏡面反射而造成日光返照，因而造成影像資訊損失，影響影像之製圖及解讀，本研究嘗試分析日光返照形成之原理，並使用能量傳輸解析與影像處理方法測試減低日光返照之影響。 讀取測試影像所得為影像灰度值，無法轉換至輻射值，即使使用Cox and Munk (1954) 模型及MODO軟體得到之傳輸量及光源輻射值，亦無法進行輻射值改正；影像拍攝當時海表面模型未知，所計算之傳輸量無法符合每個點位之差異，難以進行日光返照之改正。因此先嘗試從影像色彩不均及資訊損失之補償方法著手，使用白平衡及去霧處理兩種影像處理方法進行測試，測試結果發現日光返照區域有過度增顯情形，範圍增大且亮度誇張化。但若將日光返照區域視為亮度特徵，把影像RGB色彩空間轉為HSV色彩空間後，以亮度進行二維經驗模態分解，發現原始影像減去模擬影像之殘餘項與日光返照之範圍與特徵相當，將其視為雜訊後去除後得到之影像，可以有效改善了日光返照之影響。 使用平均值、均方根誤差、峰值訊號雜訊比及結構相似性指標對於影像結果進行統計指標計算，並與肉眼觀察結果進行比較，RGB波段平均值主要與原始影像色調有關，SSIM指標值可以看出色彩結構的維持，但主要以人眼判識為主。

When taking aerial photography images on sea area at daytime, the sea surface reflection of solar radiation cause the sun glint. It will cause the loss of image information, even continuous images stitching. In this paper bring up the definition of sun glint, and use the radiation transfer analysis and image processing methods for testing. We readed images to get digital numbers, but we can’t transfer it to radiance.Even we use Cox and Munk (1954) Model and MODO to get the transmittance and solar radiance, we still can’t correct the radiance.There was no models of sea surface when shooting images,the transmittance we calculated can’t apply for every point of surface, it can’t be used to correct sun glint. For the uneven image colors and the loss of image information, using white balancing and dehazing for testing, the results found sun glint region increased significantly, increase the range and brightness exaggerated. If sun glint area can be regarded as the brightness characteristics and allowed images from RGB color space into HSV color space, the Value processed with Bidimensional Empirical Mode Decomposition(BEMD), and found that the region and characteristics of residual with sun glint of origin images are the same, and removed the residual. That results can found that BEMD really improve the sun glint. The Mean value, Root Mean Square Error (RMSE) , Peak Signal to Noise Ratio(PSNR), and Structural Similarity (SSIM), which are the statistical indicators calculated for results, and these indicators compared with the naked eyes, and finally make conclusions and suggestions.