基于稀疏表示与字典学习的图像融合方法分析.docx

摘要
图像融合是在信息融合技术的理论基础上发展起来的图像处理技术,是把来自多个不同传感器或者同一传感器在不同时刻的同一场景的多幅图像融合成一幅图像,来获取对场景的更为准确清晰的表示,使之更加适合人类的视觉感知和后续的图像处理。然而,随着信息需求量的日益增加,信号的带宽越来越宽,在信息获取中对采样速率的要求越来越高,因此导致了数据泛滥。传统的图像融合方法随着数据量的不断增大,给信号的存储和传输带来了巨大的压力。奈奎斯特采样理论无法满足信息的获取与传输,压缩感知理论所提出的新的数据获取方法为之提供了新的思路。
本文是在传统的图像融合框架的基础上,结合信号处理及其应用领域中广泛应用的结构——稀疏性,提出了一种基于离散小波变换和稀疏表示的图像融合方法,通过离散小波变换将图像分解为不稀疏的低频部分和可以近似看做稀疏的高频部分,然后将高频部分进行稀疏表示,分别采用不同的融合规则进行融合,最后将融合后的低频部分和高频部分经过离散小波逆变换,得到最终的融合图像。通过对医学图像,多聚焦图像和红外—可见光图像这三种不同类型的图像进行融合,来验证本文方法的有效性。
在图像的稀疏表示中,字典的选择具有很重要的影响。本文比较了两种不同的过完备字典对图像融合结果的影响,一种是固定字典,该类字典的选择取决于字典能够在多大程度上稀疏表示信号;另一种是通过 KSVD 方法来训练得到的字典,该类字典可以通过选择初始训练字典来使得训练出的字典能够对信号进行很好的稀疏表示。另外 KSVD 字典的学****过程具有除噪的性能,所以能够从带有噪声的图像中训练出干净的字典。
关键字: 图像融合 稀疏表示 字典学****KSVD
Abstract
Image fusion is a new image processing technology which arises from the theory of information fusion, it fuses many images into a single image, e from different sensors or one sensor at different time for the same scene, to get a more accurate and clear description for human visual perception and subsequent image processing task. However, with the increasing demand for information, the bandwidth of signals e more and more wide, and requires a higher demand for sampling in the information acquisition, thus leading to the data flood. With the amount of data continuously increasing, the traditional image fusion method provides a huge pressure for the storage and transmission of signals. Nyquist sampling theory fails to meet the demands, the new data acquisition ing pressive sensing provides a new train of thought for it.
This paper proposes a new image fusion method based on DWT and sparse representation on the base of traditional image fusion framework, combiningthe widely
used structure in the signal processing and its application—sparsity. The proposed
method poses the image into the low ponents that are not sparse and the high ponents that are approximately sparse, and then make sparse representation for high ponents, using different fusion rules for the high pon