本論文提出一個適合在手持式影像擷取裝置上使用的文件影像二值化的系統，它可以處理因環境光源、低品質文件或有汙損的文件影像，使得後續的文字擷取與辨識能夠順利的進行。二值化的處理流程包含影像區塊數量的決定，以及區塊影像二值化的決定。在區塊數量的決定上，我們使用一個有效偵測影像紋理的方法，並根據紋理特性抽取出相應的特徵值，接下來我們將這些特徵以類神經網路學習並決定影像的區塊數量。在影像區塊二值化的決定，我們根據區塊的變異程度來決定特徵，其次為了找出影像區塊的特性，我們透過背景評估的方式計算每一個區塊的差異並以此為特徵，接著我們使用模糊理論的方法，首先決定輸入與輸出的模糊集合，接著建立相對應的模糊規則，最後以解模糊化方法來決定影像的二值化門檻值。本論文所提出的文件影像二值化研究方法，能夠有效的解決因手持式擷取裝置所造成的，亮度非均勻、低對比或陰影等因外在環境所造成文件影像辨識不佳的結果，以及能夠有效的抵抗區域二值化後所造成的雜訊問題。
This paper proposes a new adaptive document image binarization algorithm for hand-held camera. This method can solve non-uniform illuminant problem. It is divided into two parts: the determination of block number of an image and the threshold value of block image. First, we will divide image into many equal-sized regions with texture features and artificial neural network. The Laws’ mask and sobel edge detector method are used to extract an image texture features. And then, these features are inputted into neural network. The learning algorithm of neural network uses the error back-propagation learning algorithm. Subsequently, the three features are extracted from each region. Finally, we use fuzzy inference method to determine the threshold value for each region. Tests on images produced under uniform and non-uniform illumination conditions show that our proposed method yields better visual quality and better OCR performance than three locally adaptive binarization methods.
