【作者】 孙尚白；

【导师】 董军宇；

【作者基本信息】 中国海洋大学 ， 计算机应用技术， 2010， 硕士

【摘要】 近年来,超分辨率已成为图像处理领域中的研究热点。所谓图像超分辨率处理就是从一序列降质的低分辨率图像中获取高分辨率的图像。超分辨率技术已经广泛应用在卫星遥感、军事侦察、医学诊断等很多领域。目前三维表面纹理超分辨率技术的研究是图像处理领域的前沿方向,有着广阔的研究前景和现实意义。小波分析是一种时频分析工具,它在图像处理领域有重要应用。小波分析用于超分辨率是当前研究的一个热点。小波变换具有多分辨率特性和时频局部化特性,但它只有有限的方向表示,不能很好地表征图像中的方向信息。Contourlet变换是在小波变换的基础上发展而来的,它继承了小波变换的优点,而且具有多方向性和各向异性,能够更有效地捕捉图像中的边缘信息。Contourlet变换在图像超分辨率处理方面具有广阔的应用前景。本论文在前期三维表面纹理超分辨率方法的基础上,结合小波变换和Contourlet变换,提出了小波超分辨率法和Contourlet超分辨率法。在小波超分辨率法中,针对具有相似纹理结构的多类三维表面纹理,应用小波变换,并结合textons替代方法,对三维表面纹理进行超分辨处理。文章用多组实验数据证明了该方法的一定有效性。为克服小波变换的不足,本文进而使用Contourlet系数表示图像特征,采用Contourlet变换域系数关系,构建统一的训练集,并结合支持向量机方法,发展了更具通用性和有效性的Contourlet超分辨率算法。最后,本文对不同超分辨方法的结果图像进行了评价研究。实验表明,Contourlet超分辨率法的结果图像具有较高的信息熵和清晰度,更好地保留了图像的纹理信息,且有较好的视觉效果。更多还原

【Abstract】 Image super-resolution has become an active research area. It refers to as image processing algorithms to reconstruct high-resolution(HR) images. Super-resolution technology has been widely used in remote sensing, military surveillance and medical diagnosis, etc. Currently, super resolution technique of 3D surface textures is an less studied research field.As a time-frequency analysis tool, wavelet analysis is important in image processing applications. Wavelet analysis for super-resolution is an active research topic in the field. Wavelet transform has good properties such as multi-resolution and time frequency localization, but it has only limited direction that cannot capture the direction information of the image properly. Contourlet transform is developed based on wavelet transform, and it not only inherits the advantages of wavelet transform, but has multi directional and anisotropy. Therefore it is promising to image super-resolution processing since it can capture significant information from a natural image.Based on the previous super resolution methods for 3D surface textures, this paper proposes two new super resolution methods. One is based on wavelet transform and developed for many types of 3D surface textures with similar texture, by using wavelet transform and textons substitution. Finally a set of experiments and results are produced to show the evidences of the efficiency of the algorithm.To overcome the weaknesses of wavelet transform, the other algorithm is proposed based on contourlet transform. It has a general training set by studying the correlation of contourlet coefficients and combining the SVM theories. This algorithm makes the super resolution of 3D surface textures easier and more practical. Finally, the super-resolution results produced by different methods are evaluated. This new algorithm can effectively maintain the edges information and outperforms other methods in terms of entropy, definition and visual qualities.更多还原