结合高斯多尺度变换和颜色复杂度计算的显著区域检测

提出一种结合全局和局部对比度的数据驱动自底向上的视觉计算模型,采用高斯多尺度变换和颜色复杂度计算相结合的方法检测图像的显著区域。图像经过高斯金子塔多尺度变换后,用改进的冗余计算方法,在频域中消除全局冗余信息,初步将大范围的目标从周围环境中分离出来,并均匀地突显目标,利用多尺度特征子图点对点平方融合使全局显著区域显著度进一步增强,背景冗余信息进一步得到消弱;同时采用改进的像素颜色复杂度计算方法得到局部显著图,最后用最大类间方差法(Ostu)分割出图像中的显著区域。在多种自然图像上进行实验,实验结果表明,该方法能迅速检测出与人类视觉结果一致的显著区域。

Salient region detection based on Gaussian multi-scale transform and color complexity measure

A bottom-up data driven visual attention computational model based on global and local contrast is proposed in this paper,which combines Gaussian multi-scale transform and color complexity measure to detect salient region.After Gaussian pyramid multi-scale transform,global redundancy information is removed with improved redundancy measure method in frequency domain;the large-scale object is primarily separated from its surroundings and uniformly highlighted.And square incorporation of multiple scale feature sub-maps can further decrease the redundancy information and improve the contrast between salient and non-salient regions.In the mean time,local salient map can be obtained using multi-scale analysis of colour complexity.Finally,salient region can be extracted using Ostu algorithm.This method was tested on various natural images,and experiment results show that this approach can quickly extract salient regions consistent with human visual perception.