利用局部可適性法則改良經驗模態分解法以去除不良光影; A modified EMD method using the local adaptation strategy and its application to non-uniform illumination removal

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/48500

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/48500

Title:	利用局部可適性法則改良經驗模態分解法以去除不良光影;A modified EMD method using the local adaptation strategy and its application to non-uniform illumination removal
Authors:	葉宜益;Yi-yi Yeh
Contributors:	資訊工程研究所
Keywords:	經驗模態分解法;不良光影;彩色影像;局部可適性法則;最小平方近似法;EMD;non-uniform illumination;color image;AEMD;least-squares approximation
Date:	2011-08-16
Issue Date:	2012-01-05 14:56:29 (UTC+8)
Abstract:	本論文利用經驗模態分解法(Empirical Mode Decomposition , EMD)原理，提出一個影像光場估測與矯正的方法AEMD(Adaptive EMD)以解決相機取像影像上的不良光照影響；與EMD不同的是本論文不需要IMF訊號的抽取與下包絡線(Min-Envelope)和均值包絡線(Mean-Envelope)計算，只需考慮上包絡線(Max-Envelope)計算得到光場描述。本系統主要分成三個部分：1.原始影像光場運算、2. 調整光場描述、3.光影移除。　第一部分:我們首先以一維的方式對輸入灰階值訊號，以最小平方近似法求得上包絡線。第二部分：經由上包絡線的分佈與特徵找出原始訊號光場差最大的點當做分段點並分割原始訊號，之後分別對每段訊號重新做最小平方近似法運算，求得新的局部可適性上包絡線，也就是較佳的光場描述。第三部分:以光場強度來設定對比度並強化灰階值訊號，根據不同光場類型求取影像調整對比度，之後調整灰階值做光影的移除。　實驗證明，本論文的方法對於曝光過度、陰影、反光與透光的文件影像處理都有不錯的效果，在處理的時間上不僅遠遠勝過EMD，且對文字辨識也可達到與傳統的EMD方法相同的效果。　另外針對彩色影像，可適性包絡線的求取改採以三角內插法(Triangle Linear Interpolation)來完成，並對色彩空間中亮度(value)分量進行調整，以保留彩色連續性特質，實驗結果對不良光影移除亦有時顯著的效果。　In this thesis, we focus on solving the problem of uneven light effect on document images. A method based on EMD theorem is proposed to estimate and rectify the light-field. Our proposed method does not need to extract the IMF signal and only need to consider the max-envelope instead of all envelopes and residue function in EMD to obtain the light-field distributions of images. Our system consists of three parts including max-envelope calculation, light-field distribution adjustment, and signal scale adjustment. 　In the first part, we first use fast method, least-squares approximation, to find max-envelope form input signal scale via one dimensional way. In the second part, we observe the distributions of max-envelope and signal to find the max differential point between light-fields as the cut points. Then, segment the signal by the cut points and calculate local max-envelopes to obtain closer light-field distribution of original signal. In the third part, we will set the contrast value depending on light-field intensity and enhance the gray scale signal. The contrast value will be fixed if it is an over-exposure signal. Last, enhance gray scale signal by contrast value and finally adjust the enhanced gray scale. 　Experimental results show that our method exhibits good outcome about removing over-exposure, shadow, reflection and back-light effects on document images. Besides, the execution time is much faster than EMD and the recognition result can achieve the same efficiency of EMD method. 　As to the color image uneven light-field effect, we calculate the max-envelope by “Triangle Linear Interpolation”, and adjust the value-element (V of HSV color system) to conserve the property of continuous color information. Experimental results demonstrate that our proposed method performs well on color images too.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	673	View/Open

社群 sharing

Loading...