图像理解中的视觉感知与图像的关联组织

Visual Perception and Image Association Architecture in Image Understanding

对注意力焦点附近的图像片进行视觉关联学习,建立视觉特征与语义特征之间的感知联系,使图像的视觉感知行为和图像理解的认知行为连成一个整体,模拟人类的认知行为过程.自底向上的特征映射和自顶向下的语义判定相结合,建立了图像与语义类别之间的基础框架联系.在用户对查询结果的反馈中,保留正相关的图像剔除负相关图像,动态地实现了不同查询要求下图像数据的重组.实验表明,在基础框架之下,多层次语义群组关联和多框架协同能够较好地复用了原有框架组织下的结果,有效地实现了视觉感知、语义理解和查询案例之间的联系,满足了图像柔性检索的需要.

In visual contact learning of image blocks near the focus of attention, the perception contact of visual feature and semantic feature of image blocks is built up, making visual perception behavior of image and cognitive behavior in image understanding into a whole, which imitates the human cognitive process. Combination of bottom up feature map with top down semantic predications es- tablishes the basic framework contact between images and their semantic classes. In user feedback to query results, positive correlation images are reserved and the negative correlation images are removed. Experiment shows that, under the basic frameworks, through multi-level semantic group link and multi-framework coordination, the images of original framework are well reused, and the correla- tion between the perception, semantic understanding and query cases is effectively realized, which meets the requirement of flexible retrieval of images.