基于连续一致性集的鲁棒多模型拟合算法研究-计算机科学与技术专业论文.docxVIP

优秀毕业论文 精品参考文献资料 AbstractA Abstract A large number of computer vision applications are dependent on the information of interest extracted from the image．In the computer vision problem，the information usually refers to the parameters of the model that describe the image，and we can use the model fitting algorithm to estimate the parameters of the model from the observed data．The model fitting algorithm plays an important role in large number of computer vision problems．In computer vision，the parameter estimation has been deeply studied．However，due to the outliers and multiple structures that are unavoidable present，traditional model fitting algorithms do not deal well with these issues．Therefore，how to efficiently recover from the contaminated visual data is still a challenging problem． Based on the extensive research and summarization of the relevant literature on the field，we analyze and discuss several important problems for multi-strucutre model fitting．Followed by several recent representative robust model fitting，we first review the current development of the robust model fitting．Then we extend the traditional consistent set to the continuous space from the discrete space，we propose a new continuous consistency which can more accurately describe the consistency of information．A new more robust model fitting framework based on continuous and consistent set is presented．To deal with the robust multi—model fitting algorithm in the presence of imbalanced data present in the data，we propose a robust model fitting method called Structure Decision Graph fitting algorithms．In order to deal with the situation that it is hard to judge two hypotheses whether they from the same structure if the priori information is not used．We introduce the conception of relation networks，and propose a new model fitting framework based on relation networks．Finally,the model fitting for the future development of robust model fitting algorithms is given． III 万方数据 Abstract Experimental Abstr