高级检索
    鲍复民, 李爱国, 覃 征. 基于SGNN的图像融合[J]. 计算机研究与发展, 2005, 42(3).
    引用本文: 鲍复民, 李爱国, 覃 征. 基于SGNN的图像融合[J]. 计算机研究与发展, 2005, 42(3).
    Bao Fumin, Li Aiguo, Qin Zheng. Image Fusion Using SGNN[J]. Journal of Computer Research and Development, 2005, 42(3).
    Citation: Bao Fumin, Li Aiguo, Qin Zheng. Image Fusion Using SGNN[J]. Journal of Computer Research and Development, 2005, 42(3).

    基于SGNN的图像融合

    Image Fusion Using SGNN

    • 摘要: 近年来,将神经网络用于图像融合处理取得了一些成果,但已有的方法存在着计算量大、需要用户设置网络结构和较多参数等缺点.自生成神经网络(SGNN)是一类自组织神经网络,它不需要用户指定网络结构和学习参数,而且不需要迭代学习,是一类特点突出的神经网络.提出一种基于SGNN进行图像融合的新方法,分3步:①对图像进行预处理,使用小波方法滤除图像的噪声;②用SGNN对图像像素进行聚类,将像素按灰度值聚为某几类;③对经过第2步处理的像素进行融合,用灰度值对像素进行模糊分类之后再用加权平均法精确化,最终得到融合图像.该方法易于使用、计算速度快.实验表明该方法融合结果的均方误差比拉普拉斯金字塔算法和小波变换方法降低约30%~60%.

       

      Abstract: There have been some reports on approaches to image fusion based on neural networks recently. But these approaches have some shortages, such as being complex on computing, and having to have their network structures and many parameters to be set by users in advance. Self-generating neural network (SGNN) is a self-organization neural network, whose network structures and parameters need not to be set by users, and its learning process needs no iteration. In this paper, an approach to image fusion using SGNN is proposed. This newly proposed approach consists of three steps: ①pre-processing of the images. It removes the noises in the images by discrete wavelet transforms; ②clustering pixels using SGNN; and ③ fusing images using fuzzy logic algorithms. The approach has advantages of being wieldy used by users and having high computing efficiency. The experimental results demonstrate that the MSE (mean square error) reduced by this proposed approach decreases 30%~60% than that by Laplacian pyramid and discrete wavelet transform approaches.

       

    /

    返回文章
    返回