基于视觉感知的图像显著区域的提取

Extraction of salient region based on visual perception

基于Itti模型,提出了一种改进的模型来提取图像显著区域,采用Itti方法提取图像的亮度、朝向特征显著图,在此基础上,将图像的频域特征融入到图像的颜色特征提取中,并且加入图像的轮廊特征提取,避免了Itti模型提取特征时没有明显的轮廊边界的现象。在显著图的合并阶段,采用局部迭代法取代直接相加的合并方式。此模型与Itti模型相比,提取的显著图效果更加明显。

Based on the Itti model, this paper puts forward an improved model to extract the image salient region, extract the brightness map and orientation map by using Itti method. On the basis of this, the color feature is obtained in frequency domain, and the image contour feature extraction is joined in order to avoid the phenomenon of no obvious contour boundary in Itti model. In the merge phase of saliency map, it uses the local iteration methed instead of the directly adding method. Compared the model with Itti model, the saliency map effect is more obvious.