基于稀疏表示的自适应图像融合方法研究

STUDY ON SPARSE REPRESENTATION BASED ADAPTIVE IMAGE FUSION METHODS

图像融合是将两幅或两幅以上由不同传感器在同一时间或不同时间获取的关于某个具体场景的图像或图像序列信息融合为一幅图像,使得融合图像更有利于人们分析和理解。基于目前经典的图像融合方法的基础,提出一种新的基于稀疏表示的自适应图像融合算法。首先根据训练的超完备字典将两幅源图像表示为两组稀疏系数,然后根据系数的特征自适应地选取融合规则对系数进行融合,最后由融合系数和字典进行重构得到融合图像。该算法在稀疏表示的过程中能够有效地避免产生块效应且能去除噪声,从而提高图像质量。实验结果表明该方法在主观和客观评价上均优于其他算法。

Image fusion means combining two or more images or image sequence information about one particular scene taken by different sensors either simultaneously or unsimultaneously, so that the fused image is more analyzable and comprehensible. On the basis of present classical image fusion methods, the paper proposes a new sparse representation based adaptive image fusion algorithm. The method firstly presents two source images as two groups of sparse coefficients by trained over-completeddictionary; then according to coefficient characteristics adaptively chooses fusion rules to fuse coefficients; finally reconstructs fused coefficients and dictionary to obtain the fused image. During sparse representation process the algorithm can effectively avoid the generation of block-effect and can eliminate noises; therefore the image quality is improved. Experimental results show that the proposed method is superior to the other algorithms either at subjective or objective evaluations.