视觉注意模型在显著图生成中的适用性评估

摘要

Aiming at the problem of application scope of ltti-Koch, Stentiford. Spectral Residual and Hu-Rajan-Chia model, this paper evaluates four saliency maps generated by the four visual attention models. It combines with eye movement experiment, and calculates respectively their similarity of the four saliency maps by the four models and eye movement data. Experimental results shew that the four models have their own characteristics and lay emphasis on different aspects, and the similarity between saliency maps by models and saliency map by eye movement data is no less than 0.7.%针对Itti-Koch、Stentiford、光谱剩余假说和Hu-Rajan-Chia4种视觉注意模型适用范围的问题,通过4种模型生成的显著图,结合眼动实验分别计算显著图间的相似度,由此对4种模型的适用性进行评估.实验结果表明,4种视觉注意模型对包含不同特征和语义的图像各有侧重,通过眼动数据得到显著图的相似度不低于0.7.