應用稀疏表示演算法於衛星影像全色態銳化之研究

Remote Sensing Image Pan-sharpening using Sparse Representation

郭湘琳(Hsiang-Lin Kuo)

指導教授 ： 徐百輝

國立臺灣大學/工學院/土木工程學研究所/碩士(2017年)

https://doi.org/10.6342/NTU201703630

摘要

全色態銳化為一種結合全色態影像的高空間解析度與多光譜影像的多光譜優勢，最終獲得高解析度多光譜影像的衛星影像融合技術。過去已有許多全色態銳化相關的研究被提出，然而多種演算法各有其不同的限制與缺失，如IHS轉換法的光譜失真與小波轉換法的空間幾何變形等問題。為了獲得高解析度多光譜影像，全色態銳化亦可視為是一種提升多光譜影像解析度的技術。近年來在影像處理中，出現一種新的影像解析度提升技術，稱為影像超解析，可有效提升靜態影像或動態視訊之解析度。其中，稀疏表示技術是與壓縮感知相關的研究議題，這幾年成功並廣泛地應用於影像處理的反轉換應用，如去除雜訊、影像復原以及影像超解析等。稀疏表示為一能將信號以少量且顯著之字典以及稀疏係數之線性組合來表示，且能恢復或重建原始訊號之技術，可有效解決一般影像超解析恢復資訊過少的問題。
本研究為應用稀疏表示的原理以及影像字典訓練而提出的影像融合法。高解析度的字典來自Pan影像；低解析度字典則來自降解析度至與MS模糊程度相同的Pan影像。稀疏係數為字典排列方式，由正交匹配追蹤演算法計算而得。依據稀疏表示的重建特性，經由替換事先訓練好的高、低字典對，將全色態影像的紋理注入多光譜影像中，達成影像融合的目的。與多種現有全色態銳化法比較之實驗成果顯示，本研究提出的應用稀疏表示的全色態銳化法，不僅恢復全色態影像完整精確的空間細節，同時保持更豐富的原始多光譜影像的光譜資訊，成功避免光譜失真及幾何變形等影像融合問題。

關鍵字

全色態銳化 ； 影像融合 ； 影像超解析 ； 稀疏表示 ； 字典學習

並列摘要

Pan-sharpening is a fusion technique of synthesizing a single high spectral and spatial resolution MS image from a low-resolution MS image and a high-resolution PAN image. Considerable research has been devoted to remote image fusing technique during the past years. However, traditional methods like intensity hue saturation suffers from the problem of spectral distortion. And recent methods like discrete wavelet transform may cause spatial distortion. In order to create a high-resolution MS from a low-resolution MS, Pan-sharpening can be defined as an image restoration problem. Recently, a new image restoration technique called Super-Resolution (SR) has attracted much interest and has been applied into many image processing areas. The goal of SR methods is to recover a high resolution image/video from one or more low resolution input images/videos. Although SR is an ill-posed problem, making precise recovery impossible, sparse representation demonstrates both effectiveness and robustness in regularizing the inverse problem. Sparse Representation theory supposes any input signal x can be represented as a linear combination, i.e., x=Dα. The vector α is called the sparse representation coefficient of x with over-complete dictionary D.
In this paper, a novel remote sensing image fusion method is proposed with sparse representations over learned dictionaries. The high-resolution and low-resolution dictionaries are learned from the source Pan images and degraded Pan images respectively. The sparse coefficients of the Pan image and low-resolution Pan image are sought by the orthogonal matching pursuit algorithm. Then, the fused high-resolution MS image can be recovered by combining the obtained sparse coefficients and the high-resolution dictionary. The quantitative results and visual evaluation on the WorldView-2 data show that the proposed method is comparable or even superior to the existing pan-sharpening methods. The proposed method not only preserve spectral and spatial details of the source images but overcoming the drawbacks of fusion distortion.

並列關鍵字

Pan-sharpening ； Image fusion ； Super resolution ； Sparse representation ； Dictionary learning

參考文獻

鄭心惠，2002。遙測影像空間品質之評估，碩士論文，國立成功大學，台南市，台灣，101頁。

邱彥瑋，2012。混合式多光譜影像全色態銳化方法之探討，碩士論文，國立台灣大學，台北市，台灣，58 頁。

Aharon, M., Elad, M., and Bruckstein, A., 2006. K-SVD: An Algorithm For Designing Overcomplete Dictionaries for Sparse Representation. IEEE Transaction on Signal Process. 54: 4311–4322.

Amro, I., Mateos, J., Vega, M., Molina, R. and Katsaggelos, A.K., 2011. A survey of classical methods and new trends in pan-sharpening of multispectral images, EURASIP Journal on Advances in Signal Processing, Vol. 2011, no. 79, pp. 1-22.

Chavez P. S. Jr., Sides, S. C., and Anderson, J. A., 1991. Comparison of Three Different Methods to Merge Multiresolution and Multispectral Data: Landsat TM and SPOT Panchromatic. Photogrammetric Engineering & Remote Sensing, 57(3):265-303.