基于视觉认知的图像搜索与识别关键技术研究

发布时间：2018-04-29 22:17

本文选题：图像搜索 + 视觉认知　；参考：《北京科技大学》2015年博士论文

【摘要】：图像模式识别的主要挑战在于如图像检索和理解等不断变化的高层次处理要求、难于表达的图像内容以及图像表达的数宇阵列与通常可以被人类所接受的概念化内容之间的语义鸿沟。本文在视觉认知获得图像视觉信息的基础上,将视觉搜索与认知阶段的语义判定等高层处理联系起来,使图像区域或图像块与图像语义和图像类别之间建立感知联系,采用神经索引方法把感知到的兴趣内容进行索引和存储并通过认知粗糙集的处理实现了认知规律的发现,在任务事件的联合驱动下完成对图像的协同搜索和识别。通过动眼对图像区域进行扫描和搜索,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,使图像的视觉认知行为和图像识别的认知行为连成一个整体,使图像识别融入了视觉认知和语义解释两个方面,符合人类的认知规律。图像搜索时的特征映射与语义判定相结合,形成图像语义类别之问的基础框架联系。在语义关联的基础上形成场景类别的语义表达,便于用户将感知上相似的图像组织在一起,形成概念上下文,用户可以解释和标记图像而无需给出图像的概念描述。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。在基础框架辅助下,多层次语义群组关联和多框架协同较好地复用了原有框架组织下的结果,有效地实现了视觉认知、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。
[Abstract]:The main challenge of image pattern recognition is the changing high-level processing requirements such as image retrieval and understanding. The semantic gap between the image content which is difficult to express and the digital array of image expression and the conceptualized content that is generally accepted by human beings. On the basis of obtaining image visual information by visual cognition, this paper links visual search with high-level processing such as semantic decision in cognitive stage to establish perceptual connection between image region or image block, image semantics and image category. The perceptual content of interest is indexed and stored by neural index method, and the discovery of cognitive laws is realized by processing cognitive rough sets, and the cooperative search and recognition of images are completed under the joint driving of task events. By scanning and searching the image region through eye movement, the visual association learning of the image film near the focus of attention is carried out, and the perceptual relationship between the visual feature and the semantic feature is established. The visual cognitive behavior of the image and the cognitive behavior of the image recognition are combined into a whole, and the image recognition is integrated into the visual cognition and the semantic interpretation, which accords with the cognitive law of human beings. The feature mapping in image search is combined with semantic decision to form the basic framework of image semantic category. The semantic representation of scene categories is formed on the basis of semantic association. It is convenient for users to organize perceptually similar images together and form conceptual context. Users can interpret and mark images without giving the conceptual description of images. In the feedback of the user to the query result, the positive correlation image is retained to remove the negative correlation image, and the recombination of the image data under different query requirements is realized dynamically. With the help of the basic framework, multi-level semantic group association and multi-frame collaboration can reuse the results of the original framework, and realize the connection between visual cognition, semantic understanding and query cases. It meets the need of flexible image retrieval.
【学位授予单位】：北京科技大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TP391.41

【参考文献】