高度光線變化影響之影像的分析及處理技術開發

Abstract

隨著數位取像裝置(digital capture device)的技術日漸進步與使用者的日益增加，功能的完善和價格便宜成為購買的重要條件。一般在攝影時使用者最關注的三個方面分別是光線、焦距和色彩，所以對於自動曝光補償、自動對焦和自動白平衡得功能要求也相對的提高，故本論文針對光線所引起的影像問題發展出新的解決技術，首先，我們提出了一種文件影像二值化的方法，此方法會依據光線分佈的強弱來決定適當的門檻值，我們利用同質性(Homogeneity)的方法將影像分為兩個區域，再利用影像處理技術中的中值濾波器(Median filter)來得到最後的光線分佈的強弱影像，依據此影像的指示配合我們所提出來的適應性門檻值決定的演算法來做二值化，其所得到的結果與其他的方法比較可以發現我們所提出之方法能夠避免光線的影響而得到一個完整的二值化影像。接下來我們提出了一個逆光影像偵測與補償的演算法，逆光補償(Backlight compensation)是許多數位取像裝置中都會提供的一個功能，但是其補償的效果不理想並且未有偵測的功能存在，有鑑於此我們從影像的空間位置(Spatial position)和亮度分佈機率統計圖(Histogram)抽取出兩個指示影像逆光程度的指標，利用模糊推論(Fuzzy inference)的方法整合這兩個指標來完成逆光偵測。在逆光補償方面，我們提出利用曲線來調整影像的亮度，此曲線可以提升逆光影像中主體的亮度，並讓影像中的其它部份的亮度維持住，如此便能完成補償的動作，在此我們使用兩個數學方程式來描述這個曲線，一個是拋物線方程式(Parabolic equation)，另一個是三次方程式(Cubic equation)，從實驗的結果顯示使用三次方程式為補償曲線的結果比使用拋物線方程式的結果好，因為它有更平滑的特性。

This dissertation is aimed at the image problems with varying lighting effects. First, we have a new method of document Image binarization, and with this method we can decide the proper threshold value by the distribution of light, and we use this method of the homogeneity to separate the images into two parts, and then use one of the technology approaches, the Median filter, to get the very last image of light distribution. According to its directives, we match up with the algorithm decided by our proper threshold value to get a threshold, and try to compare the result with the one of the other methods, we found that our method can avoid the influence of light and has got a complete binarized image. Subsequently, we proposed an algorithm of backlight image detection and compensation. Backlight compensation is the function which many digital capture devices would provide, but the effects of the compensation are not good enough to what we have expected, and even there was not the function of detection. Seeing that we choose two indicators of directing image backlight levels from the image spatial position and its histogram, by using the fuzzy inference, we integrate the two indicators to complete the backlight detection. In the backlight compensation, we suggest that using it to adjust the brightness of images because the curve can elevate the brightness of the main object, and keep the brightness of other parts. By that, we can complete the compensation. Herein, we use two formulas to describe this curve. One is the parabolic equation; the other is the cubic equation. From the results of the experiment, using the cubic equation to be the calculation of the compensation curve is better than using the parabolic one, because it has more flexible characteristic.