本研究主要是提出一套影像量測系統,用於半導體封裝後之規格檢測。目前許多半導體製程仍須使用量測儀器或是人工目視來檢視生產後之產品規格,而購買使用量測儀器不僅需要較大的花費,且須定時對儀器做保養以及校正,而人工目視也需受過專業訓練以及成熟度有緊密相關,這些都是需要花費較大的金錢、時間、人力等成本,因此建構此套系統不僅可以改善上述的成本問題,還可減少量測的複雜度,以及減少量測的人力與時間等效益。 影像量測系統主要使用下列幾項方法;影像前處理部分,會先將原影像轉換成灰階影像,接著利用平均濾波器(Average Filter) 去除及降低影像中的雜訊干擾,以利後續的特徵擷取與減少誤判。接下來膨脹、侵蝕、斷開、閉合部分則是形態學上抽取影像成份的工具,主要是表示區域形狀的描述,特別是邊界、骨架都是有用的,而研究中就是需要這個特點把區域形狀描繪凸顯出來。接著再利用 Sobel運算子分割出邊界,Sobel運算子主要在處理影像分割問題,在技術上它是一離散性差分算子,用來運算圖像亮度函數的梯度之近似值,可分割出梯度不同的相鄰圖像。最後Hough Transform可以用來檢測任意形狀的曲線,依照屬性將分割的圖像用線條描繪出來,並做最後的距離量測。
This study presents an image identification and measurement system for examining and testing the packaged semiconductor specifications. Most semiconductor processes rely on measurement instruments or inspectors to examine the finished product specifications visually. Measurement gauges are costly and have to be maintained and calibrated regularly; inspectors have to be trained professionally and their performance depends on the maturity of their skills, which requires enormous costs, time, and efforts. Therefore, an automatic identification and measurement system will not only reduce costs, but minimize the complexity of measurement substantially and reduce the manpower and time needed for measurement. Image identification and measurement system incorporates following steps. The treatment begins with image pre-treatment procedure to transform original images into gray scale images, followed by average filter to remove and reduce noises from the images to facilitate feature extraction and minimize miscarriage. In the next step, images are extracted by morphological tools, such as dilation, erosion, opening and closing to portray the shape of area, especially the boundaries and framework. This shape of area is highlighted by the features stated above. In the next step, Sobel operator is employed to sever the boundary. Sobel operator, a discrete difference operator designed to sever boundaries, is an ideal tool to process the issues related to dividing images and to calculate the gradient approximation of brightness function and break up the images next to each other with different gradients. Finally, Hough transform is employed to examine the curves of any shape and depict the severed images by lines and measure distance in accordance with their attributes.