摘要： 为在没有先验知识的情况下准确获取图像显著性目标，提出一种基于对数Gabor滤波器和超复数傅里叶变换的视觉显著性检测算法。利用对数Gabor滤波器模仿人类视觉感受野，对输入图像进行预处理，提取颜色、纹理方向等特征。根据所得特征构造各尺度下的超复数图像，并求其傅里叶变换相位谱，将多尺度超复数相位谱反变换后进行归一化，从而获得视觉显著图。实验结果表明，该算法与传统的算法相比具有更高的准确率，应用于复杂场景下的交通标志检测能取得较好的检测效果。

关键词: 视觉显著性, 对数Gabor, 超复数, 傅里叶变换, 多尺度, 显著图

Abstract: In order to obtain more accurate salient object from an image in the absence of priori knowledge, this paper proposes a visual saliency detection algorithm based on Log-Gabor filter and hypercomplex Fourier transform. It uses Log-Gabor filter to process input image and obtain color and texture feature, constructs a hypercomplex image using feature images, and calculates its Fourier transform phase spectral. It calculates visual saliency map by normalization. Experimental results show that the proposed method outperforms state-of-the-art methods remarkably in visual saliency and has better detection results in traffic sign detection.

Key words: visual saliency, Log-Gabor, hypercomplex, Fourier transform, multi-scale, saliency map

中图分类号: 

• TP312