應用橢圓霍氏轉換量測重疊葉片面積之影像處理方法

An Image Processing Method to Measure Overlapped Leaf Area Using Elliptical Hough Transform

錢中方(Chung-Fang Chien) ； 林達德(Ta-Te Lin)

《農業機械學刊》 9卷4期 (2000/12) Pp. 47-64

https://doi.org/10.30062/JAM.200012.0004

摘要

種苗葉片面積為量測其生長狀態的重要資料，應用影像處理方法可以達到非破壞量測葉片面積之目的。由於許多蔬菜種苗的葉片形狀是接近橢圓形或卵形，因此用橢圓形的方式來概略描述這些葉片是一種可能的方法。在本研究中以甘藍菜、青花菜及白菜種苗為實驗材料，於擷取種苗葉片影像後，應用橢圓霍氏轉換來找出最近似種苗葉片的橢圓形及位置，而為了降低計算的複雜度與記憶體需求，在本研究中發展了一種新的演算法，先將影像解析度由512×512降至32×32像素，在其中初步找出橢圓的位置及大小後，再逐步升高影像的解析度，並在前一階段已得知橢圓位置附近，繼續搜尋提高解析度後橢圓之正確位置，最後再據以推估葉片面積。在本研究中發現以所對應的橢圓來描述甘藍菜、青花菜與白菜種苗葉片，其葉片真實面積與橢圓面積均有相當高的線性關係，各組實驗數據線性迴歸之決定係數R^(2)均在0.95以上。以橢圓面積對葉片真實面積之迴歸式來推測預估甘藍菜、青花菜與白菜種苗葉片面積，其平均絕對誤差百分比分別為17.3±14.7%、7.5±4.5%及7.8±9.6%。所發展的演算法亦可以應用於部分重疊葉片面積的推估，兩片葉的重疊越少偵測效果越好，重疊率在40%以下時，軸向及側向重疊的面積平均絕對誤差約在±10%之內。在重疊葉片的情形下，葉片側向重疊比率在40%以內，軸向重疊比率在60%以內時，多能夠成功地辨識出個別葉片。

關鍵字

影像處理 ； 霍氏轉換 ； 橢圓 ； 葉片面積

並列摘要

Seedling leaf area is an important index in measuring its growth. Non-destructive measurement of leaf area can be achieved using image processing approach. Because the leaf shapes of many kinds of vegetable seedlings are nearly elliptical or egg shaped, it is possible to use an ellipse to approximately describe a leaf. In this study, we used Hough transform to search for ellipses by matching the boundary of seedling leaves of cabbage, Chinese cabbage and broccoli separately. In order to reduce the computing complexity and memory requirement of elliptical Hough transform, a new algorithm was implemented beginning with reduction of image resolution from 512×512 to 32×32 pixels. With the low-resolution image, an ellipse was initially searched with its position and dimensions determined. Using this preliminary information, the resolution of the image was stepwisely increased and corresponding ellipse with higher resolution was further determined by searching the neighborhood of the ellipse determined in the previous step. The ellipse obtained at the final resolution could be used to estimate the corresponding leaf area. In this study, the area of ellipse was found to be highly correlated with leaf area (R^2＞0.95). Using the area of ellipse to estimate cabbage, broccoli and Chinese cabbage seedling leaf area, the average absolute percent errors were 17.3±14.7%, 7.5±4.5% and 7.8±9.6%, respectively. The elliptical Hough transform algorithm was also tested to estimate the total area of overlapped leaves. The less the overlapping ratio, the higher the accuracy of estimation was achieved. When the overlapping area ratio was less than 40%, the estimation errors were under ±10% for the cases of leaves overlapped in axial or lateral directions. In most cases, when the overlapping area ratios were below 40% in lateral direction or 60% in axial direction, individual leaves were successfully recognized with this algorithm.

並列關鍵字

Image processing ； Hough transform ； Ellipse ； Leaf area