基于生物激励计算模型在图像显著性提取中的研究

Research on biologically-inspired computional model for image saliency detection

提出一种生物激励的显著性特征计算模型。首先通过注意块学习从眼动数据库中选择与视觉响应一致的稀疏基;然后基于稀疏基表达原理对图像建立计算模型并提取显著性特征:包括全局连续性、区域颜色对比以及局部复杂度对比特征;再仿照细胞调节原理,提出新的特征组合方法进行特征融合。最后将该算法在多个典型的场景中对感兴趣区进行提取实验,证明比其他算法具有优越性。并提出将此算法应用于虚拟与现实场景融合中,能良好地提取出真实场景中的有效区域和剔除虚景区域。

A biologically-inspired model for the computer vision community was proposed. At first, a set of basis functions that accorded with visual responses to natural stimuli was learned by using eye-fixation patches from an eye-tracking dataset. Then image calculation model was established and features was derived based on the principle of sparse representation： including global continuity, regional color contrast, and local complexity contrast. And then refer to the principle that activity in cells responding to stimuli, a new feature combination theory was proposed to achieve features fusion. Afterwards, some experiments extracting regions of interest from typical scenes prove that this algorithm has superiontity than other algorithms, and the algorithm was applied in virtual and reality interactivity. It can effectively extract effective regions and eliminate virtual scene area.