適應性非線性擴散模式於瑕疵偵測與影像復原之應用

Adaptive anisotropic diffusion models for defect detection and image restoration

趙新民(Shin-Min Chao)

指導教授 ： 蔡篤銘

元智大學/工程學院/工業工程與管理學系/博士(2008年)

https://doi.org/10.6838/YZU.2008.00108

摘要

改善影像品質以提高人類對影像內容的理解，以及處理影像資料使機器能自動認知影像內涵，是數位影像處理的兩個主要應用領域。本論文提出四種改良式之非線性擴散模式，可同時對影像進行適應性之平滑處理及邊緣保留/強化處理，並將所提出之方法應用在影像復原及低對比影像之瑕疵偵測。

本論文所提之前兩種非線性擴散模式，在擴散係數函數中同時考量影像梯度及灰階變異數兩個因子以進行擴散處理，可有效的濾除雜訊並同時保留住影像中細微資訊，這兩種方法可適用於影像復原之應用。本論文提出的第三種與第四種方法主要應用之對象為低對比影像之瑕疵偵測，其中第三種方法在傳統的非線性擴散模式中加入了銳化的概念，所提出之非線性擴散模式可在濾除雜訊及背景紋路的同時強化細微瑕疵，而本論文的第四種方法提出了一個廣義的擴散係數函數，可藉由調整擴散係數函數之參數值來控制擴散處理之效果，同時本研究也提出了自動選取最佳參數組合之方法，使得第四種非線性擴散模式在進行瑕疵偵測時具有更佳之效率及偵測效果。

實驗結果顯示本論文所提出的前兩種方法應用在醫學影像、古字畫影像及天文影像之影像復原問題，皆能夠有效改善影像品質，提供良好的視覺效果。本論文所提的後兩種方法用於檢測具有低對比特性之背光板、玻璃基板及增亮膜等液晶顯示面板關鍵零組件之應用上，均具有良好之偵測效果。

關鍵字

影像復原 ； 瑕疵偵測 ； 液晶顯示面板 ； 邊緣保留平滑處理 ； 非線性擴散

並列摘要

Improving pictorial information for human interpretation and processing image data for autonomous machine perception are two major application directions of digital image processing. In this research, four adaptive smoothing models with edge-preservation/edge-sharpness based on anisotropic diffusion are presented for the applications of image restoration and defect detection in low-contrast surface images. The traditional anisotropic diffusion model only provides an adaptive smoothing operation based on sole gradient information. It cannot effectively preserve fine details with lower gradient or sharpen edges in a low-contrast image.

The first two diffusion models incorporate both local gradient and gray-level variance to adjust the weights of diffusion. They can effectively remove noise while preserving edges and fine details in the image. These two improved diffusion models can be well applied to image restoration for human interpretation. The third diffusion model incorporates the sharpening operator in the traditional diffusion model so that it not only performs the smoothing process for noise and background textures but also actively provides the sharpening process for object edges. The third proposed model can effectively filter out the background noise in a faultless region, yet well sharpen the anomalies in the diffused image. It can be applied to automatic defect detection in low-contrast images. The fourth diffusion model further incorporates a generalized diffusion coefficient function that uses a linear-logarithmic function for flexibly adjusting the function curve. By controlling the parameter values of the diffusion coefficient function, the diffusion model can produce more effective and efficient diffusion results for defect detection in low-contrast images. Experimental results have shown that the edge-preserving diffusion models can well improve image quality for visual interpretation in medical images, artwork images and astronomical images, and the edge-sharpening models can well detect low-contrast defects in backlight panels, glass substrates and brightness enhancement films found in liquid crystal display (LCD) manufacturing.

並列關鍵字

Image restoration ； Defect detection ； LCD inspection ； Edge-preserving smoothing ； Anisotropic diffusion

參考文獻

Aja-Fernandez, S. and C. Alberola-Lopez, 2006, On the estimation of the coefficient of variation for anisotropic diffusion speckle filtering, IEEE Transactions on Image Processing, 15, pp. 2694-2701.

Alvarez, L., P. L. Lions and J. M. Morel, 1992, Image selective smoothing and edge detection by nonlinear diffusion, SIAM J. Numerical Analysis, 29, pp. 845-866.

Bakalexis, S. A., Y. S. Boutalis and B. G. Mertzios, 2002, Edge detection and image segmentation based on nonlinear anisotropic diffusion, IEEE International Conference on Digital Signal Processing, 2, pp. 1203-1206.

Barash, D., 2002, Fundamental relationship between bilateral filtering, adaptive smoothing, and the nonlinear diffusion equation, IEEE Transactions on Pattern Analysis and Machine Intelligence, 24, pp. 844-847.

Bovik, A. C. T., S. Huang and D. C. Munson, 1987, The effect of median filtering on edge estimation and detection, IEEE Transactions on Pattern Analysis and Machine Intelligence, 9, pp. 181-194.