项目名称: 生物可信性频域视觉注意模型及其图像多语义快速获取方法研究
项目编号: No.U1304607
项目类型: 联合基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 刘尚旺
作者单位: 河南师范大学
项目金额: 30万元
中文摘要: 频域视觉注意模型是视觉计算中拟合人类视觉系统的研究热点。针对该类模型的生物视觉理论和图像多语义快速获取研究薄弱的问题,本项目在前期研究的基础上拟构建一个生物可信性四元数图像频域视觉注意模型,并结合同样有生物学背景、图像处理性能良好的脉冲耦合神经网络(PCNN)进行图像多语义快速获取。首先,基于生物侧抑制和谱白化等理论,进行图像四元数系数的选择和频域中保留相位谱的幅度谱归一化,从而建立一个生物可信性频域视觉注意模型。其次,将所建模型与PCNN整合,并利用返回抑制机制进行多目标图像分割;利用PCNN的时间签名和四元数的互相关等特征进行图像多语义快速获取。最后,拟建立一个实时图像语义检索原型系统,对本项目的方法进行验证与评价。研究成果不仅可提高计算机获取复杂图像语义的准确性和泛化性能,丰富图像理解与识别理论,而且在计算机视觉、计算机图形学和机器导航等领域有广阔应用前景。
中文关键词: 频域视觉注意模型;生物可信性;HFT模型;图像分类;目标跟踪
英文摘要: Frequency-domain visual attention model is a hot research topic in the visual computation field for mimicking human visual system. Aiming at solving the problems of biological vision theory of these models and image multi-semantic rapid acquisition, on the basis of pre-studies, this project will build a biologically plausible quaternion-image frequency-domain visual attention model, and acquire image multi-semantic rapidly by combining the model with pulse-coupled neural network (PCNN) which also has biological background and good image processing ability. Firstly, based on the theories of biological lateral inhibition and spectral whitening, this project will solve the problems of the choice of weighted quaternion components and the normalization of amplitude spectrum with its phase spectrum unchanged in frequency domain. Thus, the biologically plausible quaternion-image frequency-domain visual attention model can be built. Secondly, this project will integrate the built model with PCNN, and use inhibition of return mechanism to segment multi-objects image; utilize PCNN image signature and quaternion correlation to acquire image multi-semantic rapidly. Finally, a real-time prototype system for image semantic retrieval will be built to test and evaluate the method of this project. The research results not only c
英文关键词: frequency-domain visual attention model;biological plausibility;HFT model;image classification;target tracking