基于全局结构相似度度量方法的显著性检测

Saliency detection based on global structure similarity metrics

视觉显著性检测是计算机进行图像处理的重要步骤,它包括目标识别、图像分割等,在网络图形、指纹识别等诸多领域都获得了广泛的应用。提出了一种全局结构相似度度量方法进行显著目标检测:利用全局结构相似度对人类视觉系统的高层次抽象,并且考虑了人的视觉特性和心理感受,把全局结构相似度理论用于显著程度的度量,度量结果进行加权后求均值得到显著图,并采用二值法提取显著目标。与经典Itti算法相比,不仅克服了马赛克现象,而且考虑了全局空间信息对结构相似度的影响,能够根据计算能力和精度选定参数。

Saliency detection is an important step image processing with computer, which includes the target recognition, image segmentation, and it is widely used in many fields such as network graph- ics and fingerprint recognition. In this paper, we proposed a global approach of structural similarity measure to significant target detection： using structural similarity to the human visual system＇s high- level abstraction, and taking into account the human visual characteristics and psychological feelings. We used the theory of structural similarity to measure the significant level of the target object, and get the saliency map after weight averaging the results. At last, the method of threshold was used to extract the salient target. Compared with the classic Itti algorithm, our approach can not only overcome the mo- saic and take into account the global spatial information on the impact of structural similarity, but also select the parameters according to computing power and precision.