Abstract: Because the traditional image segmentation algorithms need the prior information of the object of interest, they usually fail to segment satisfactorily in blind images. In order to obtain the salient regions without prior, this paper proposes a salient image segmentation algorithm based on the human visual attention mechanism. Firstly, the global Gist feature is extracted to collect the correlative image set of the target image. Secondly, the salient region of the target image is computed and extracted in both target image and correlative image set by combining scale invariant feature SURF (speeded up robust features) and Lab color model space feature. Finally, the salient object can be segmented with the assistance of the efficient graph-based image segmentation. The experimental results show that the proposed method is applicable to the blind image which is visual salient semantic.

Key words: saliency, speeded up robust features (SURF), Gist, graph theory

摘要： 传统的图像分割算法在分割前需要输入目标的先验信息，因此不适应盲图像的分割。为此，提出了一种基于显著性的图像分割算法，主要借鉴人的注意力机制将图像中具有视觉显著性的区域分割出来。首先，利用Gist全局特征获取目标图像的相似图像集；然后，结合尺度不变特征SURF（speeded up robust features）和Lab颜色模型空间特征对目标图像内以及相似图像集提取显著性特征，并根据显著性块频率低的原理进行显著性分割；最后，结合图分割获得最终的显著性区域分割结果。实验结果表明该方法适用于具有显著性视觉语义的盲图像。

关键词: 显著性, SURF, Gist, 图论