基于自相似性的多聚焦图像融合

Multi-focus image fusion based on self-similarity

为更好保留源图像的信息,提出基于内容相关性的自相似区域划分技术,根据内容相关性的自相似性方法完成图像区域划分,利用图像间的共同相似区域完成图像匹配。实验中,经小波变换后的低频系数采用自相似性区域划分技术,获得区域划分标签图,采用加权融合方法完成融合;高频系数结合标签图,采用梯度能量绝对值最大值法完成融合。实验结果表明,在主客观评价上,该方法取得了较优的融合结果,符合人类的视觉感知,其性能优于经典的融合方法。

To better preserve the significant information of source images,the region segmentation technology based on content-correlated self-similarity algorithm was proposed.According to content relevance of image,the self-similarity algorithm was achieved to complete the region segmentation of image.The shared similar regions between images were used to match the images.After wavelet transforming,the self-similarity region segmentation technology was implemented in the low-frequency coefficients,the regional label map was obtained to match the source images.The low-frequency coefficients were fused through the weighted decision-making method.According to the label map of the adaptive region,the maximum value of gradient energy was used to fuse the high-frequency coefficients.Experimental results show that the proposed method achieves better results in the subjective and objective evaluation criteria,which is in accordance with the human visual perception.And its performance is superior to the classical fusion method.