臺灣古蹟、歷史建築修復現場花崗岩非破壞性調查法之研究—以影像處理法分析主要礦物比例

A Non-Destructive Investigation Method of Granite in Taiwan’s Cultural Heritage Sites: Image Analysis for Calculating Mineral Proportions in Granites

臺灣古蹟、歷史建築中有相當多由石材建造的建築和碑碣，當這些文化資產需要修復時，選擇與原物件相匹配的替換材料對於保護文化真實性至關重要。然而，對石材原件和替換材進行定量比較以往需要破壞性分析，但文化資產通常禁止採樣。為了解決這個問題，本文研發了一項非破壞性影像處理程序，得出花崗岩主要礦物的比例，以定量比較其外觀。在建築石材中，選擇花崗岩來開發這種影像處理程序法，因為它具有高辨識度顏色的礦物，並且是一種廣泛應用於文化資產的建築材料。因此，礦物比例是描述花崗岩影像的主要參數，從外觀上看，花崗岩可分為灰白色系和粉棕色系兩大類，再藉由拉曼光譜儀證實影像中顏色與各主要礦物的對應關係。在影像處理程序中，利用紅色、綠色和藍色（R-G-B）色彩空間模型中，粉紅、棕色像素與灰、白色（斜長石）和黑色（黑雲母）像素R值（紅色）的差異性，並利用2R/（G+B）– δ（RGB）關係圖分辨圖形序列中的淺色和深色像素，最後應用灰階來解析序列中的白色和灰色像素，最終將影像中顏色的像素數量轉換為礦物比例。影像處理所得之數據再與需要破壞性程序取得的理想礦物比例與進行比較，兩者誤差在10%以內。因此，影像處理程序被驗證為一種可有效辨識文化資產中花崗岩的非破壞性調查方法。最後，將此方法實際應用於「湯德章紀念公園」與「南鯤鯓代天府」兩處文化資產並進行相關討論，證實影像處理法可有效提升現場調查之效能與廣度及深度，是非常值得繼續研究開發及推廣之調查方法。

 

Heritage sites often comprise buildings and monuments constructed from stone materials. When these heritage sites require restoration, selecting replacement materials that match originals is crucial for preserving cultural authenticity. Nevertheless, sampling is generally prohibited at heritage sites, preventing destructive analyses for quantitatively comparing originals and replacements. To address this issue, we established a non-destructive image analysis protocol to derive proportions of constituent minerals for quantitatively comparing the appearance of stone materials. Among the construction stones, granite was selected for developing this color differentiation algorithm because it is a ubiquitous construction material and it comprises minerals with distinctive colors. Mineral proportions are, therefore, prime parameters for describing the images of granites. By appearance, granites were sorted into the gray-white group and the brown-pink group. The corresponding relationship between the colors in the image and the main minerals is confirmed by a Raman spectrometer. In image analysis, the brown and pink pixels were distinct from the gray, white (plagioclase), and black (biotite) ones for higher R (red) values in the red, green, and blue (RGB) color model. Moreover, the dark colored pixels had a standard deviation from the mean of RGB values [δ (RGB)] extending to higher values. Granite pixels were, therefore, first separated using a 2R/(G+B) versus δ (RGB) plot. The brown and pink pixels formed a high R array with a positive 2R/(G+B) – δ (RGB) slope, whereas the gray, white, and black pixels defined a low R array showing an inverse 2R/(G+B)–δ(RGB) correlation. The combination of δ (RGB) and (R+G+B) further separated light and dark colored pixels in a 2R/(G+B) – δ (RGB) array. At last, grayscale was applied to resolve the white and gray pixels in a low R array after isolating the black pixels. The numbers of pixels for each color were then converted to mineral proportions. More precise mineral proportions were acquired for comparison through composition 2 analyses that required destructive procedures. The comparator values were derived by solving a matrix equation that described a bulk composition as the sum of the products of the constituent mineral compositions and their corresponding proportions. The mineral proportions from image analysis and composition analyses were within 10%. Hence, our new protocol was validated as a method for selecting proper replacement materials for restoring granite-based cultural heritage sites. Finally, the image analysis procedures were applied to the restoration of two monument in southern Taiwan.