【作者】 李姗琦； 孙永宣； 高隽； 谢昭；

【Author】 Li Shanqi Sun Yongxuan Gao Jun Xie Zhao (School of Computer & Information,HeFei University of Technology,Hefei 230009,China)

【机构】 合肥工业大学计算机与信息学院；

【摘要】 图像目标分析与识别在计算机视觉领域中起着重要的作用,是近年来的研究热点之一。在分析过程中如何有效快速的识别出目标是图像处理的核心任务。本文采用融入了位置信息的层次Dirichlet过程模型进行单目标识别,在模型的训练过程中考虑目标类别间的关系,通过共享目标的部分节省了大量计算。同参数图模型相比,非参图模型即层次Dirichlet过程模型还可以根据图像数据集的规模自动学习出部分的个数,使得目标识别的结果更为精确。更多还原

【Abstract】 Currently object analysis and recognition is the central issue in the field of computer vision.In the process of image analysis,how to quickly recognize object in an effective way is a core task.This thesis achieved single object recognition based on hierarchical dirichlet Process model which add in object position information.This model considered relationships among object categories during training.This approach leads to significant computational savings by sharing parts.Compared with parametric model,nonparametric model-the hierarchical dirichlet process allows the number of latent parts to be learned automatically according to the size of the image dataset,it makes the results of object recognition more accurate.更多还原