【摘要】：數(shù)字圖像的超分辨率分析可以廣泛應(yīng)用于航空航天、精確制導(dǎo)與精確打擊、農(nóng)作物疾病診斷、自然災(zāi)害預(yù)報、身份識別與認證等諸多領(lǐng)域。根據(jù)低分辨率圖像的來源類型,可以將數(shù)字圖像的超分辨率分析分為兩類:一類是多幅圖像(或者稱為序列圖像)的超分辨率算法,另一類是單幅圖像的超分辨率算法。本文研究單幅圖像的超分辨率算法。從數(shù)字圖像超分辨率的基本概念出發(fā),將低分辨率需要內(nèi)插的圖像分為光滑部分和非光滑部分,針對兩個部分研究了單幅數(shù)字圖像的超分辨率算法。第一個部分針對光滑連續(xù)函數(shù)模型,探討了空域內(nèi)插的方法,主要包括:最近鄰內(nèi)插法、雙線性內(nèi)插法、雙三次內(nèi)插法、雙三次樣條法和二維直接內(nèi)插法。實驗結(jié)果表明,二維直接內(nèi)插法明顯優(yōu)于前面四種內(nèi)插算法,在相同復(fù)雜度的情況下,可以得到最優(yōu)的超分辨率結(jié)果;而最近鄰法在所有情況下表現(xiàn)出最差的性能,出現(xiàn)最嚴(yán)重的馬賽克現(xiàn)象;雙三次內(nèi)插法和雙三次樣條法則優(yōu)于最近鄰法和雙線性內(nèi)插法。第二個部分的內(nèi)插算法,采用了復(fù)數(shù)輪廓波變換。到目前為止,復(fù)數(shù)輪廓波變換主要有兩種類型,即固定冗余度復(fù)數(shù)輪廓波變換和可變?nèi)哂喽葟?fù)數(shù)輪廓波變換。固定冗余度復(fù)數(shù)輪廓波變換是雙樹復(fù)數(shù)輪廓波變換,可以認為是基本輪廓波變換的實數(shù)部分和虛數(shù)部分;而冗余度可變的復(fù)數(shù)輪廓波變換則是基于映射的架構(gòu),可以根據(jù)情況選取不同的冗余度。由于后者具有較高的靈活性和重組性質(zhì),論文中選取其作為變換工具。文章闡述了這種復(fù)數(shù)輪廓波變換的實現(xiàn)方法,探討了該種變換的基本性質(zhì),特別重點討論了該變換的移不變水平,在該變換的基礎(chǔ)上,設(shè)計了一種性能良好的單幅圖像內(nèi)插算法。該算法充分應(yīng)用了復(fù)數(shù)輪廓波變換的移不變性質(zhì),將低分辨率需要內(nèi)插圖像的非光滑部分,在復(fù)數(shù)輪廓波域,對變換域進行內(nèi)插,然后進行反變換,并在此基礎(chǔ)上,將本算法和二維直接內(nèi)插算法融合,得到比較滿意的實驗結(jié)果。實驗結(jié)果表明,所提出的算法能夠更好的實現(xiàn)單幅圖像的超分辨率重建,在客觀參數(shù)(例如峰值信噪比等)和主觀效果兩個方面,都有比較良好的表現(xiàn)。
[Abstract]:Super-resolution analysis of digital images can be widely used in aerospace, precision guidance and precision strike, crop disease diagnosis, natural disaster prediction, identification and authentication, and so on. According to the source type of low-resolution image, the super-resolution analysis of digital image can be divided into two categories: one is super-resolution algorithm for multiple images (or sequence image), the other is super-resolution algorithm for single image. In this paper, the super-resolution algorithm of single image is studied. Based on the basic concept of super-resolution of digital image, the image needed to be interpolated with low resolution is divided into smooth part and non-smooth part. The super-resolution algorithm of single digital image is studied for the two parts. In the first part, for the smooth continuous function model, the spatial interpolation methods are discussed, including nearest neighbor interpolation, bilinear interpolation, bicubic spline and two-dimensional direct interpolation. The experimental results show that the two-dimensional direct interpolation algorithm is superior to the previous four interpolation algorithms, and the optimal super-resolution results can be obtained under the same complexity, while the nearest neighbor method has the worst performance in all cases. The bicubic interpolation method and the bicubic spline rule are superior to the nearest neighbor method and the bilinear interpolation method. The second part of the interpolation algorithm, using the complex contour wave transform. Up to now, there are two main types of complex contour wave transform, that is, constant redundancy complex contour wave transform and variable redundancy complex contour wave transform. The constant redundancy complex contour wave transform is a double tree complex contour wave transform, which can be considered as the real and imaginary parts of the basic contour wave transformation, while the complex contour wave transformation with variable redundancy is based on the mapping framework. Different redundancy can be selected according to the situation. Because of its high flexibility and recombination, the latter is chosen as a transformation tool in this paper. In this paper, the realization method of the complex contour wave transform is expounded, the basic properties of the transformation are discussed, especially the invariant level of the transformation is discussed, on the basis of the transformation, A single image interpolation algorithm with good performance is designed. The algorithm makes full use of the shift invariant property of complex contour wave transform. The low resolution needs to interpolate the non-smooth part of the image. In the complex contour wave domain, the transform domain is interpolated, then the transform domain is inversely transformed. The proposed algorithm is fused with the two-dimensional direct interpolation algorithm and the experimental results are satisfactory. The experimental results show that the proposed algorithm can better realize the super-resolution reconstruction of a single image, and has good performance in both objective parameters (such as peak signal-to-noise ratio) and subjective effect.
【學(xué)位授予單位】：信陽師范學(xué)院
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP391.41

【參考文獻】

相關(guān)期刊論文 前3條

1 羅敏;李輝;;低照度條件下超分辨率人臉圖像采集系統(tǒng)設(shè)計與實現(xiàn)[J];計算機測量與控制;2016年11期

2 陳新武;劉真;馬文娟;薛靜;范春鳳;;三特征Contourlet紋理圖像檢索算法[J];信陽師范學(xué)院學(xué)報(自然科學(xué)版);2015年03期

3 董江,葉梧,馮穗力;內(nèi)插法在提高電視圖像清晰度中的應(yīng)用[J];電視技術(shù);2004年02期

，

本文編號：2229378