导　　师： 曾凡仔

学科专业： 081002

授予学位： 硕士

作　　者： ;

机构地区： 湖南大学

摘　　要： 人们通过图像来获取外界的主要信息，因此对图像的处理是一个重要的研究课题。图像融合是将来自传感器的多源图像融合成一幅图像的过程。融合的图像保留了多源图像的主要信息，为图像的进一步处理，提供更加清晰、有效的信息。因此，该技术目前被广泛应用于军事、医学图像处理、远程传感等领域。随着信息技术的快速发展，人们对图像质量的要求也越来越高，势必造成图像传输数据的增大，因此图像压缩也将成为一个重要的研究方向。 本文针对目前在图像融合及图像压缩编码方法存在的一些问题，提出了基于压缩感知理论的新方法。本文的主要工作如下： 首先分别介绍了图像融合和图像压缩编码技术的研究背景及意义、发展现状，然后又对压缩感知理论（Compressed Sensing，CS）的框架进行了描述，并将压缩感知理论运用到二维图像的处理中。 其次，基于压缩感知理论提出了一种采样点少且结构简单易实现的图像融合方法。算法首先对需要处理的两幅或多幅图像进行小波变换，再分别对得到的小波系数进行稀疏处理得到稀疏矩阵，并通过系数绝对值较大法进行融合，然后对融合后的系数矩阵通过随机观测获取压缩采样，而图像恢复则是对得到的压缩采样通过求解最优化的问题得到。由于对小波系数进行了稀疏处理，故该方法可以用少量的采样点来恢复图像。实验结果表明，在相同采样点下，该方法得到的图像质量明显优于传统的系数绝对值较大法融合；在少量采样点下，采用该方法也可以使融合的图像达到较好的效果。 最后，针对矢量量化压缩速度慢、图像复原效果不理想等问题，根据图像小波分解后高频子带稀疏的特点，提出了一种基于压缩感知理论的分类量化图像编码算法。仿真结果表明，与LBG矢量量化编码算法相比，重构图像质量得到极大提升，在相似压缩比下，该算法取得了较好的效果，峰值信噪比（PSNR）平均有2~4dB的明显提高；在相似PSNR下，该算法在图像压缩方面也有很大改进。 Image processing is an important research because people sense the outsideworld mainly from image. Image fusion, as a process of combining multiple imagesfrom different sensors can provide more clear and effective image for the furtherimage processing applications. Therefore, it is widely used in military, medical imageprocessing, remote sensing and so on. With the rapid development of informationtechnology, and the raising demand for high-definition, image, it will inevitably incurthe daunting cost of image aquisition, image storage and image transmission. Thus,image compression will become an important research direction. Considering the problems in the image fusion and image compression coding,the thesis proposes the new methods based on compressed sensing. The main work areas follows: Firstly, the thesis introduces the advance of the image fusion and imagecompression coding technology, and then reviews the theoretical framework and it’sapplication of compressed sensing theory in two-dimension image. Secondly, based on compressed sensing, a new image fusion method ispresented. The new method can take reduced samples, and has the advantages ofsimple structure and easy implementation. The method decomposes two or moreoriginal image using wavelet transform first, and gets the sparse matrix by the waveletcoefficient sparse representation, and fuses the sparse matrixs with the coefficientsabsolute value maximum scheme. With randomly observed, it can receive thecompressed sample. At the fusing end, the fusion image is recovered from the reducedsamples by solving the optimization. The proposed method can construct the fusionimage with less measurements because the wavelet coefficients are spare presentation.Simulation results show the proposed method exhibits its superiority over thetraditional method of the maximum absolute values fusion with the same samplingrates, and under the lower sampling, it can also achieve better fusion performance. Finally, for the problem in vector quantization that the compression ratio is slowand image restoration result is not satisfactory, and according to the sparsityproperties of the high frequency wavelet transform coefficients, the thesis proposed anew method of categories quantization coding for image based on compressed sensing.Compared with the LBG vector quantization coding algorithm, simulation results demonstrated that the proposed algorithm improved the quality of the recoveredimage significantly. For the similar compression ratio, the peak signal to noise ratio ofthe proposed algorithm was improved about2~4dB. For the similar PSNR, thealgorithm in image compression has also improved significantly.

关 键 词： 压缩感知 小波分解 图像融合 图像压缩 分类量化编码

分 类 号： [TP391.41]

领　　域： [自动化与计算机技术] [自动化与计算机技术]