Abstract: Multi-focus image would be divided into focus area and defocus area. Focus area usually attracts attention with outstanding visual salience. So a novel image fusion method is presented for multi-focus color images in which human visual system is simulated to design the algorithm. The Hyper Fast Fourier Transformation （HFFT） is used to improve the traditional algorithm for saliency detection in order to get saliency map of a color image directly. The fixational eye movement is simulated by changing the scale of the image lightly. So that more saliency maps of an image could be obtained in different size and could be accumulated in same size. The pixels of the source image with the most saliency are selected to compose the fusion image. Experimental results show that HFFT makes full use of color information of the source images and the accumulated saliency could enhance the performance in PSNR and RMSE. More than 36 dB of PSNR, less than single indicator of RMSE and the best subjective evaluation could be obtained easily in compared experiments.

Key words: hyper Fourier transformation, saliency, fixational eye movement, multi-focus image fusion

摘要： 多聚焦图像存在聚焦和离焦区。通常聚焦区吸引人的注意力，具有突出的视觉显著性。提出一种模拟人类视觉机制的彩色图像融合方法。利用超复数傅里叶变换改进传统显著性检测算法，直接检测彩色图像显著度;模拟人类视觉的注视眼动机制对图像尺度稍微扰动，并对不同尺度下获得的图像显著度进行累加;选择累计显著度最大的源图像像素构建融合图像，并采用颜色相似性准则处理边界带像素。相关仿真实验表明，超复数傅里叶变换充分利用了源图像的颜色信息，累加显著度能提高峰值信噪比并减少均方根误差。实验结果表明该方法可得到超过36 dB的峰值信噪比和小于个位数的均方误差，在图像融合中可得到相对好的主观视觉效果。

关键词: 超复数傅里叶变换, 显著性, 注视眼动, 多聚焦图像融合