基于背景和前景交互传播的图像显著性检测

为了更精确地提取图像中的显著性区域,提出一种新的基于背景和前景交互传播的图像显著性检测计算模型。通过建立一个新的模型来寻找图像中的显著性元素,用一种交互式特征传播方法来扩散显著性特征。采用不同参数对图像进行分割,得到多个尺度下的超像素;在单一尺度下通过背景和前景交互传播获得超像素的显著值;对多个显著值进行加权平均融合,并采用平滑机制进行优化得到最终显著图。在公开图像数据库进行的试验结果表明,该模型提高了对图像显著目标大小的适应性,不仅较好地抑制了噪声,还使得显著目标更均匀地凸显出来,结果优于同类的算法。

A visual saliency detection based on background and foreground interaction

In order to extract the salient region of image efficiently, a new algorithm model of image saliency detection based on background and foreground interaction was proposed. To find the significant elements in the image, a new model using an interactive feature propagation method to diffuse the significant features was built. The image was segmented into superpixels with different parameters. The salient value of each superpixel was obtained by background and foreground interaction according to a single scale. The final saliency map was obtained by the weighted average fusion of multiple salient values in different scale, and the optimization using the smoothing mechanism. Experimental results showed that the proposed method performed better than the other state-of-the-art methods, which improved the adaptability to the size of salient regions. In addition, our method was proved better not only in restraining the noise, but also in making the salient objects more uniform.