基于多尺度张量空间的改进Itti视觉显著性检测

Improved Itti Visual Saliency Detection Based on Multi-scale Tensor Space

针对内部致密均匀且边界清晰明确图像,显著性方法检测得到显著性区域边界不精确且比较模糊,使得目标物体不连通问题,提出了基于多尺度张量空间的改进Itti视觉显著性检测算法,该方法引入张量空间特征,保存了原始图像特征的空间结构和相关性,可以很好的获取内部致密均匀图像的特征,使得目标物体连通,并结合显著性检测算法完成特征提取及目标检测。实验结果表明:检测算法提取的显著性区域结果更加接近对象实际边缘,达到更好的检测效果。

In view of internal dense uniform and clear borders image, through the saliency detection the target boundary is vague, so that the target object is not connected. In order to solve this problem, an improved Itti visual saliency detection method based on multi-scale tensor space was proposed. The method introduced the tensor space features, which preserved the original image spatial structure and correlation features, that could extract internal dense uniform image features, which made the target object connect, combining with saliency detection algorithm to finish feature extraction and target detection. Experimental results show that the proposed method can clearly and accurately extract saliency regions and achieve better detection results.