基于视觉注意和模糊区域生长的图像检索

Image retrieval based on visual attention and fuzzy regions growing

视觉注意能够反映用户对图像场景中主要目标的理解,在此基础上提出一种基于视觉注意和模糊区域生长的图像检索算法.首先,由改进的视觉注意模型得到显著图;然后,根据分割的结果,提出一种根据显著区域隶属度进行模糊区域生长的算法,合并相似区域以获得查询目标,并提取颜色和纹理特征;最后设计结合隶属度和区域邻接图的相似性度量准则.实验结果表明,该算法能够有效表达用户查询的语义,具有较好的检索性能.

Visual attention reflects custom understanding of focused object in image scene. Because of this mechanism,an image retrieval algorithm based on visual attention regions is proposed. Firstly, a saliency map is computed by improved visual attention model. Then, according to the segmentation result, a saliency region fuzzy growing algorithm based on degree of membership is proposed to obtain object region by merging similar regions,so as to extract color and texture features. Finally,a combination of degree of membership and region adjacency graphs （RAG） strategy is designed to similarity measure. Experimental results show that proposed algorithm represent custom＇s query semantic effectively,and achieve satisfying retrieval performance.