多尺度下幅度谱与相位谱相融合的视觉注意建模

针对现有大多数频域显著性检测算法仅单独使用频域幅度谱或相位谱的不足,提出了多尺度下频域幅度谱与相位谱相结合的视觉注意模型。该模型先对图像进行四元变换以得到幅度谱和相位谱,然后对幅度谱进行了伽马修正和高斯滤波,最后采用信息熵作为权重对多尺度显著图进行融合。在两个公开数据集Bruce和Judd上,采用ROC曲线、AUC值和F-Measure测量方法对算法进行了验证和评估。实验结果表明提出的算法优于现有的5种视觉注意模型,能够更准确地预测出人们注意的显著区域,取得了更令人满意的结果。

Visual Attention Modeling Based on Multi-scale Fusion of Amplitude Spectrum and Phase Spectrum

Aiming at the deficiency of most existing frequency domain saliency detection algorithms which only use frequency domain amplitude spectrum or phase spectrum information alone,a visual attention model combining multi-scale frequency amplitude spectrum and phase spectrum was proposed.Firstly,the amplitude spectrum and the phase spectrum are obtained by quaternary transforming the image,and then the gamma correction and Gaussian filtering are performed on the amplitude spectrum.Finally,the entropy is used as weight to fuse the multi-scale saliency map.On the two published data sets,Bruce and Judd,the ROC curve,AUC value and F-Measure method were used to validate and evaluate the algorithm.The experimental results show that the proposed algorithm outperforms five visual attention models, predicts the significant attention areas more accurately,and achieves more satisfactory results.