對焦輪廓重建技術

Shape Reconstruction from Focus

馮騰榔(Teng-Lang Feng)

指導教授 ： 陳傳生

元智大學/工程學院/機械工程學系/博士(2006年)

https://doi.org/10.6838/YZU.2006.00282

摘要

被動式自動對焦技術（Passive autofocus），不需額外測距裝置，將拍攝數位化影像，經由清晰度函數（Sharpness function）運算，計算出所謂清晰度尺度（Sharpness measure）。改變鏡頭與被攝物體間相對距離，判斷最大清晰度所在鏡頭位置，獲得清晰的影像。常見於數位相機的主動式自動對焦技術，必需藉助測距裝置，測量物件相對於鏡頭的位置，據以調整 CCD鏡距以達成對焦目的。被動式自動對焦技術不需要測距裝置，利用鏡頭擷取的數位化影像，計算所謂的清晰度尺度(sharpness measure)後，改變鏡頭焦距或是物距，尋找清晰度尺度達到最大值的成像位置以呈現最清晰的畫面。
回顧文獻中提出多種清晰度函數，利用差分的方法對數位化影像進行計算，對於反光度高物體（如：鋼球）和低照度環境下， Sobel和 amplitude無法作為清晰度判斷指標（Index）；利用DCT計算影像清晰度，只在分析平面物件（如：報紙）時效果較佳。FFT透過對數位化影像進行，分解和合成的濾波處理。驗證平面物件、單個立體物件、易受反光物件和低照度環境下，成功得到清晰度判斷指標。DWT必須考慮2項因素：Base Function和分解次數的選擇，將一張數位化影像分解成4個部分。分別包含影像中：低解析度部分，水平、垂直和對角線方向的係數，產生變化量大的部分。經由實驗驗證，影像中垂直方向發生變化，所產生係數具有較高的空間解析度，對於清晰度的敏銳度較高。所以本文取出垂直方向，係數變化部分的資料。此資料為二維矩陣，加總所有係數值，計算出清晰度值，作為清晰度判斷的指標。由此成計算清晰度尺度結果不正確，將搜，推廣到多段差物件分析，亦得到良好效果，證實DWT適用性廣亦具實用性。
對焦搜尋法的開發，文獻中提出Binary Search Method和Fibonacci Search Method。缺點在焦距（Focal length）附近搜索頻繁，造成鏡頭移動量過大，若拍攝數位化影像內含有雜訊成分，造尋到局部最大值（Local maximum）而非總體最大值（Global maximum）。本文為了提高對焦速度及準確性，提出 YZU焦搜尋法，利用測量值進行邏輯判斷。易於將雜訊點刪除。搜索過程中分為粗搜索和細搜索兩階段，搜索到焦距附近時，假設曲線呈二次多項式分佈，預估最大清晰度尺度，減少鏡頭移動次數並快速找到對焦點（Focus），完成自動對焦工作。
重建物體的三維形貌（3D shape recovery）。拍攝一連串不同物距的數位化影像，將每張影像分割成數個 視窗（window），計算一連串影像的清晰度值。假設焦距附近的數值分佈，為高斯函數（Gauss function）分佈型態，轉換清晰度最大值，為重建物體外貌的高度值。本文設定2個門檻值（Threshold） 和 ，作為正確高度選取的標準。若清晰度對物距分佈寬度小於 和清晰度最大值大於 ，即符合高斯函數分佈。藉助此法篩選清晰影像，得到正確高度值；而非受雜訊干擾得到假高度。下一步，取出每個 矩陣，採用2階Lagrange Polynomial Equation來近似物體輪廓，利用已知點高度值估計未知部位的高度值，還原原始影像中物體的三維形貌。我們找出一連串影像中，各部位最清晰影像，合成一張全面對焦的清晰物體影像，實踐完全聚焦三維成像技術（All-Focused Image Generation）。
研究自動對焦領域中，本文完成一套被動式自動對焦系統，並成功以DWT作為清晰度判斷指標。進而利用幾十張數位化影像，重建物體的三維形貌，並合成一張全面對焦的清晰物體影像。

關鍵字

Adaptive sharpness ； Automatic focus ； Discrete wavelet transformation

並列摘要

Recovering the shape information from a set of images sensed by a camera is an important problem in computer vision. Autofocusing algorithms mainly depend on pixel-based analysis in digital camera. A passive automatic focusing system with a charge-coupled device (CCD) camera is installed to compare the sharpness function of the best focus with different measures algorithms in this thesis. In addition, several other factors, like low illumination environment, reflective object, and overlap objects are also considered.
According to the results obtained, when distance overlaid under low illumination environment, in order to obtain the best sharpness function of the image through the a autofocus CCD camera from diffuse and specula surface object, the four different bases transform of wavelet, such as Harr, Daubechies, Coiflets and Biorthogonal, are presented in this study. Meanwhile, to produce an adaptive sharpness function of an image and show its sharpness, especially in the vicinity of focus, the Sobel edge-enhanced operator, gray-level amplitude operator (amplitude), sum-modulus-difference operator, discrete cosine transformation and discrete wavelet transform vertical (DWT) were compared further. In this thesis presents a new Yuan-Ze University (YZU) algorithm to detect the position of the sharpest image from a rough surface of an industry component quickly and accurately. To prove the performance of the YZU algorithm, it is compared to some well-known methods like fast Fourier transform (FFT), Amplitude, Laplacian, discreet cosine transform (DCT), and conventional edge operators. Moreover, a new dynamic search method that implements this algorithm to produce real time digital image systems with fast response, accuracy, and robustness is proposed.
The experiment results show that this technique is applicable to practical 3D measurement. In the application, two specimens a gauge block and an integrated circuit (IC) lead frame are tested to demonstrate the validity of 3D Shape recovery are presented.

並列關鍵字

Adaptive sharpness ； Automatic focus ； Discrete wavelet transformation

參考文獻

1. J. M. Kuo, and K. S. Chen, The Application of Wavelets Correlator for Ship Wake Detection in SAR Images, IEEE Trans. on Geoscience and Remote Sensing, Vol. 41, No. 6, 2003, pp. 1506-1511.

2. M. Penedo, W. A. Pearlman, P. G. Tahoces, M. Souto, and J. J. Vidal, Region-Based Wavelet Coding Method for Digital Mammography, IEEE Trans. on Medical Imaging, Vol. 22, No. 10, 2003, pp. 1288-1296.

5. O. Kazushige, K. Izumi, M. Nozaki. and I. Takeda, An Advanced Autofocus System for Video Camera Using Quasi Condition Reasoning, IEEE Transactions on Consumer Electronics, Vol. 36, No. 3, Aug. 1990, pp. 526-530.

6. J. Fdez-valdivia, J.A. Garcia, J. Martinez-Baena, and R. Fdez-Vidal, The Selection of Natural Scales in 2D Images Using Adaptive Gabor Filtering, IEEE Trans Pattern Anal. Machine Intell., Vol. 20, May 1998, pp. 458-469.

7. T.N. Pappas, and D.L.Neuhoff, Least-Squares Model-Based Halftoning, IEEE Trans Image Processing, Vol. 8, No.8, August 1999, pp. 1102-1116.