距离测度学习理论与应用研究.pdfVIP

摘要 图像检索系统中的相关反馈过程中，根据反馈中样本正负例个数之比自适 应调整未标注样本对距离测度学习过程的影响。该算法应用于图像检索系 统中能够有效提高系统的检索精度； 3．将基于图的半监督学习和距离测度学习过程相结合，提出了一种样本标号一 距离测度同步学习算法，在迭代过程中一方面利用学习得到的样本类别标 号指导距离测度学习，另+’方面利用学习得到的距离测度改进建图过程中 样本问的距离度量，最终达到改进半监督学习的建图过程，提高样本类别 预测准确率的目的。 关键词：距离测度学习，子审问学习，半监督学习，k近邻分类，图像与 视频分类 Abstraet ABSTRACT measurementliesattheheartof classificationand Samplesimilarity many clustering methodsin andmachine distance patternrecognition learning．Thepredefinedmetric， suchasEuclideanDistancecannot USacorrect between in give similarity samples it isaconvenientdistancemetric．New of manycases，although representations and be more distancemetricscan learnedfeature samples appropriate by can the of andclassification． transformation，which improveperformanceclustering Metric transformintoanew ifthe learningapproaches samples actually space，and dimensionalofthenew islowerthanthe reductionis one，dimension ity space original metric an rolein performedsimultaneously．Distance learningplaysimportantmany as anddocument applications，suchimage face and andSOon． imagedigitalimagerecognition Inthis focustheresearchonthe