基于空域和频域的图像显著区域检测  

Image salient region detection based on spatial and frequency domains

作　　者：纪超[1] 刘慧英[1] 孙景峰[1] 贺胜[1] 黄民主[1] 

机构地区：[1]西北工业大学自动化学院,西安710072

出　　处：《吉林大学学报（工学版）》2014年第1期177-183,共7页Journal of Jilin University:Engineering and Technology Edition

基　　金：航空科学基金项目(2012ZC53042)

摘　　要：根据人眼视觉注意机制,提出将图像在空域中采用局部复杂密度对比和全局颜色分布估计,在频域中通过有效频段分割的方法分别提取显著特征,再仿照细胞调节原理进行特征组合。局部复杂密度对比是模仿万有引力定律,通过稀疏基建模的方式计算视觉注意力的大小;提出在频域内采用有效频段分割,结合幅度信息提取显著特征后加权合成。仿真实验证明本文算法能高效地检测出场景中的显著区域。最后将本文算法应用于虚拟与现实交互中检测真实场景中的有效区,效果良好。According to the human visual attention mechanism, an image salient region detection method is proposed. The salient features of image are extracted in spatial domain using local clutter density contrast and global color distribution estimation, and extracted in frequency domain using the efficient band divided method. Then, referring to the principle that cells respond to stimuli, such features are combined using the theory of feature combination. Local clutter density contrast imitates the law of universal gravitation, computing the visual attention on sparse matrix model. The efficient band divided method is used in combining information of amplitude for salient feature detection in frequency domain, later utilizes all features with weight. Experiment results show that the algorithm can effectively extract salient regions. The visual attention model is effectively applied to virtual and reality interaction to detect the effective region in real scene.

关 键 词：计算机应用 显著特征提取 稀疏表达 频段分割 特征组合 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]