【作者】 王丽芳；

【导师】 沙莎；

【作者基本信息】 中南大学 ， 计算机科学与技术， 2011， 硕士

【摘要】 目标分类识别是图像处理和模式识别领域的重要研究课题,在场景视图和智能导航等计算机视觉方面有着广泛的应用。针对目标的特征选取和分类识别这两个关键性问题,本文在对属性进行分层排序的基础上,选择最优的特征排序组合,提出一种基于属性区分力的快速层次识别算法,该方法可在信息不完备的条件下对目标进行分类与识别。实验结果表明该方法大大提高了程序的运行效率,且满足准确性和实时性的要求。所做的具体工作和创新性研究如下：1、针对目标信息呈现的不全面、不精确现象,本文引入渐进视觉目标识别机制的思想,可以在反复多次、多角度、多时段条件下不断补充正确信息,去伪存真,从而提高对事物判读的准确度和精细度。2、论文针对机器学习和模式识别中各个特征项之间由于冗余和不相关特征的存在,增加了数据量的存储代价的问题,在研究Fisher准则和分层聚类特征选择算法的基础上,提出一种基于属性区分力的特征选择排序算法,该算法能够计算出各个属性对于任意对象组合间的区分力大小。3、论文针对目标特征不合理利用而引起的分类识别速度慢、准确性低的问题,在贝叶斯信息融合分类识别方法的基础上,提出一种基于属性区分力的快速层次目标识别算法,对属性排序后在特征不完备的情况下进行信息融合,与朴素的贝叶斯方法和一般的层次识别方法相比,该算法的识别率和准确率有很大程度的提高。4、论文通过两个典型的应用实例,在对目标层次识别过程推理的基础上,进行场景图像分割及道路区域识别,实验结果表明,本文提出的算法具有较好的性能,很大程度上提高了目标分类识别的效率,对研究各类场景中对象分类和感兴趣目标识别问题具有一定的参考价值。最后,总结全文并进一步提出后续工作思路。更多还原

【Abstract】 Object classification and identification is an important research topic in the fields of image processing and pattern recognition, and it has been widely used in computer vision like scene view and intelligent navigation. According to the two key problems of feature extraction and object classification, on the basis of layering and sorting the features, selecting the optimal feature permutation and combination, the paper proposes a fast layering decision algorithm based on attribute discriminatory power which can classify and recognize the objects under the incomplete information condition. Experimental results show that this method greatly improves the running efficiency of program and satisfies the real-time and accuracy requirements.The specific work and innovations are as follows:Firstly, according to the incomplete and inaccurate phenomenon of object information presentation, the paper introduces the thought of evolutionary vision object recognition mechanism which can supplement the correct information under the condition of multi-replication, multi-angle and multi-period and discard the false and retain the true. Thereby, the accuracy and fineness of object interpretation can be improved.Secondly, according to the existence of redundancy and irrelevance between each feature in machine learning and pattern recognition and the storage cost increment problem of data size, on the basis of researching the feature selection algorithms of Fisher criterion and hierarchical clustering, the paper proposes a feature selection and sorting algorithm, which can figure out the attribute separating capacity between any object combination.Thirdly, according to the problem of slow speed and low accuracy of classification and identification caused by unreasonable utilization of object features, on the basis of classification and recognition method of Bayes information fusion, the paper proposes a fast layering recognition algorithm based on attribute discriminatory power, which fuses the information in the case of incomplete features after sorting the attributes. Compared with the Naive Bayes and common hierarchy recognition method, the recognition rate and accuracy rate of the algorithm has been greatly improved.Fourthly, through two typical application examples, the paper segments the scene image and recognizes the road area on the basis of reasoning the process of object layering recognition. The experimental results show that the algorithm proposed in the paper has a better performance and largely improves the object classification and recognition efficiency, which will have reference value to the research of object classification in various scenes and interested target recognition problem.At last, the whole paper is summarized and further follow-up working thoughts are raised.更多还原