本研究主要是藉由彩色機器視覺技術,來檢測物體之規則紋路表面上的異常瑕疵。由於空間域的檢測技術容易受雜訊所影響,因此本研究使用頻率域之賈柏轉換法(Gabor transform)來進行瑕疵檢測。傳統賈柏轉換法是利用影像灰階值的資訊來進行紋路分析,屬於單一灰階資訊的影像處理,因此容易喪失部份重要色彩資訊而無法檢測出瑕疵。基於傳統賈柏轉換法的缺點,本研究運用頻率域之賈柏轉換法的技術,並結合色彩模型(color model)之色彩特徵(color features),進行彩色紋路之表面瑕疵檢測。 賈柏轉換法是將影像資訊從空間域轉為頻率域來擷取一個視窗(window)內的紋路特徵,藉此降低對雜訊的敏感度,賈柏轉換法為一非線性正弦(sin)函數,在轉換式?迉]含了三個參數,即頻率(frequency) 、相角(orientation) 及帶寬(bandwidth) ,以此三個參數當作紋路之特徵值。彩色賈柏轉換法是擷取色彩模型中兩個與亮度無關之色彩特徵,將兩個色彩特徵組合成一複數形式來取代灰階影像的單一灰階值,以更多的影像資訊進行彩色紋路之瑕疵檢測。實驗中以紡織品、木紋、格狀布紋及毛衣為測試樣本,由實驗結果得知,彩色賈柏轉換法可明確的顯示灰階賈柏轉換法無法凸顯之瑕疵。
In this research, we use color machine vision to detect defects in homogenous texture surfaces. In order to prevent the noise interference in the spatial domain, we employ Gabor transform method in the frequency domain to detect defects. The traditional Gabor transform method is based on gray-scale image processing, which utilizes single gray-scale information to analyze textures, thus it is fairly easy to lose important color information and fails in defect detection. In this research the Gabor transform method in the frequency domain is incorporated with two color features derived from color spaces to detect defects in colored texture surfaces. Gabor transform converts the image information from spatial domain into frequency domain, and represents the texture features in a sliding window to reduce the noise interference. Basically, Gabor transform method is a non-linear sinusoidal function. It contains three parameters in the transform process, namely, frequency, orientation and bandwidth, and uses these three parameters as the texture features. In this research, two brightness-invariant color features obtained from color models are used to form a complex number, which replaces single gray-scale in the gray image, for colored texture representation. The proposed Color Gabor transform convolutes the two-color-feature complex number with the Gabor filter in a sliding window. A homogeneous region will generate zero-energy response in the convoluted image, whereas a defective region will yield large energy response. Experimented on knitted cloth, wood, checkered cloth and weave have shown that the proposed Color Gabor transform can accurately detect the defect which traditional Gabor transform can not perform in gray-scale images.