摘要人的视觉注意力机制，使得人能够从大量复杂的信息中地找到所感兴趣的少量目标信 息。自底向上的图像的显著性分析，在基于人的视觉注意力的基础上，利用图像内容本身 对于人眼的刺激，从各种背景不同的图像中，快速且相对准确地提取出目标物体，从而大 大提高计算机的效率。本文研究的是基于傅立叶变换的图像显著性分析。第一部分介绍傅 立叶变换的数学原理及物理意义，研究了傅立叶变化在图像显著性检测方面的作用。第二 部分研究了基于傅立叶变换的经典检测算法：基于谱残差的算法（SR）。由于 SR 在检测中 效果不佳，因此分析频谱残差所代表的意义。基于此，首先设计实验说明图像的振幅谱和 相位谱在图像中的作用，以此说明谱残差在显著性检测中的“无用性”。为了进一步论证这 一猜想，研究基于相位谱的傅立叶变换算法（PFT）。在 PFT 中，直接使谱残差不参与计 算，观察得到的实验结果。文章的最后一部分研究了基于超矩阵的傅立叶变换算法 HFT。 HFT 算法创建了一系列的谱尺度空间，每个谱尺度对应着一张显著性图，根据每张图的熵 值选择一张作为最终的显著图。由于 HFT 算法最终只选择了其中的一张显著图，其他的 7 张图都闲置了，所以在第四部分的第二小节研究了元胞自动机理论。元胞自动机能够根据 局部规则，将多幅图像作为输入图像，不同图像的同一位置上的像素视为邻居，将各个图 像进行融合。本文利用元胞自动机理论，将 HFT 的 8 张不同谱尺度的显著图作为输入图像 进行融合，充分地利用了每张图。78753

毕业论文关键词 显著性分析 傅立叶变换 频谱残差 超傅立叶变换 元胞自动机

Title Analysis of Significance of Image Based on Fourier Transform

Abstract Human visual attention mechanism, which make people can find a small  amount of target information they are interested in from a large number of complex information in place。 The bottom-up significant analysis of the image, which based on the human visual attention ,make use of the irritation to the human eye of image content itself ,can extract the target object quickly and relatively accurately from a variety of different and complex background images, thus greatly improving the efficiency of the computer。 The test is a significant image analysis Fourier Transform。 Firstly, introduces the mathematical principle and the physical meaning of the Fourier transform, then, we research the role Fourier transform plays in the saliency detection of image。 In the second part, we design the SR algorithm 。 SR algorithm is classic saliency detection algorithm, which is based on the concept of residual spectrum。 Experimental show that the result of SR were not good as expected。 For this, we analyze the significance of spectrum residual represents。 Firstly, design experimental to know the role of the amplitude spectrum and phase spectrum of image, as an example of spectral residuals saliency detection in the 'uselessness'。 To further demonstrate 'uselessness' conjecture, design PFT algorithm based on the Fourier transform of phase spectrum。 In the PFT, make the spectrum residuals do not participate in the calculation directly, observe the experimental 。 The last part of the test research the HFT algorithm which based on Super Fourier matrix。 HFT algorithm creates a series of spectral spatial scales, each scale spectrum corresponds to a saliency map, according to the entropy of each graph as the final choice

of a saliency map。 Because HFT algorithm ultimately select only one of a saliency map, the other seven plans are idle, so in the second part of the fourth section presents cellular automata theory。 Based on the local rules， cellular automata take the multiple images as an input image, pixels in the same position on different images are considered neighbors, make the respective image fusion。 In the paper, to make use of all the 8 saliency map, the cellular automata is used to fuse the 8 different scales saliency map of HFT algorithm。