【正文】 圖像超分辨率重建技術(shù)研究 摘 要 超分辨率圖像重建是指將一系列各自包含一部分不同圖像細(xì)節(jié)的相似低分辨率圖像通過一定方法融合成一幅高分辨率圖像的處理方法。對(duì)于一定的數(shù)字成像系統(tǒng)，通過超分辨率圖像重建 可以得到高質(zhì)量的圖像而不需要提高系統(tǒng)的硬件性能，因此它可以被廣泛應(yīng)用于圖像處理的各種領(lǐng)域中。 超分辨率圖像重建算法可以分為兩大類，即頻域算法和空域算法。頻域算法是早期的超分辨率 重建算法，只能對(duì)全局位移的圖像序列進(jìn)行處理，空域算法使用通用的觀察模型，具有更好的適應(yīng)性和重構(gòu)效果，是目前的主要研究方向。其中 POCS 算法利用投影至凸集的原理進(jìn)行圖像重建，算法直觀而且有效，是最有前途的超分辨率重建算法之一。 在序列圖像超分辨率重建算法中，本文對(duì) POCS 方法做了比較細(xì)致的研究，針對(duì)用常規(guī) POCS 算法重建的高分辨率圖像中出現(xiàn)的導(dǎo)致圖像邊緣質(zhì)量下降的現(xiàn)象 ,分析了圖像邊緣模糊形成的原因，對(duì)算法進(jìn)行了改進(jìn)，提出了邊緣保持的 POCS 算法，用基于梯度的插值算法來獲取 POCS 的初始值，實(shí)驗(yàn)結(jié)果表明，該方法能夠得到較好的視覺效果，明顯地提高重建圖像的邊緣質(zhì)量。 關(guān)鍵詞 ： 超分辨率復(fù)原，凸集投影，邊緣保持 ABSTRACT Superresolution image restoration is a method that bines multiple similar but not identical lowresolution images into a higher resolution single image. In a particular digital imaging system,using superresolution methods can help to obtain higher quality images without up grading system hard ware,therefore superresolution techniques can be applied to vast imageprocessing areas. Superresolution techniques can be divided into two main categories:frequency domain methods and spatial domain methods. Frequency domain methods are earlier superresolution methods,they can only deal with image sequences that only translational motions are allowed. Spatial methods use general observation models,they have better adaptability and performance. In POCS method,projection onto convex sets theories is employed to realize superresolution restoration,it is intuitive in theoretical and has good reconstruction performance. POCS is one of the most promising superresolution restoration methods. In sequence LR images reconstruction,the research work in this paper focuses on POCS,analyzing the factors which influence the results of super resolution. According to the traditional superresolution image reconstruction algorithm led to the blur edge of reconstructed high image. With an analysis of the reason which led to the blur edge of highresolution reconstruction image,a new image interpolation based on gradient is proposed,it kept large information of the image edge. Use this new image interpolation to get the original value of the POCS. The experimental tests show that this algorithm can significantly improve the quality of the reconstruction image. Key words： superresolution restore, projection onto convex sets, edge maintain 目 錄 中文摘要 ............................................................................................Ⅰ ABSTRACT........................................................................................Ⅱ 1 緒論 .................................................................................................. 1 研究背景及研究意義 ........................................................................... 1 圖像超分辨率重建的應(yīng)用 ..................................................................... 2 本論文的主要研究?jī)?nèi)容及其完成的工作 .................................................... 3 2 超分辨率圖像重 建 算法研究 ............................................................ 4 概述 .............................................................................................. 4 頻域方法 ......................................................................................... 5 空域方法 ......................................................................................... 6 非均勻空間樣本插值算法 ............................................................... 6 迭代反投影算法 .......................................................................... 7 凸集投影算法 ............................................................................. 7 最大后驗(yàn)概率估計(jì)和最大似然估計(jì)算法 .............................................. 7 混合 MAP/POCS 方法 ...................................................................... 8 常見的圖像重構(gòu)算法的比較 ................................................................... 8 3 POCS 算法及其改進(jìn)算法的實(shí)現(xiàn) ...................................................... 11 凸集投影算法 ................................................................................... 11 用插值方法構(gòu)造參考幀 ..................................................................12 運(yùn)動(dòng)估計(jì) ..................................................................................12 基于點(diǎn)擴(kuò)散函數(shù)的修正 .................................................................................. 13 POCS 圖像重建算法步驟 ................................................................................ 13 保留邊緣信息的 POCS 超分辨率圖像重建算法 .............................................14 4 試驗(yàn)結(jié)果與分析 .............................................................................16 試驗(yàn)低分辨率圖像的生成 .....................................................................16 改進(jìn)的 POCS 算法重建 結(jié)果及其分析 ........................................................17 5 結(jié)論 ................................................................................................18 論文總結(jié) ........................................................................................18 工作 展望 ........................................................................................18 參考文獻(xiàn) ............................................................................................20 1 1 緒論