A new distance measure for non-identical data with application to image classification

详细信息    查看全文

• 作者：Muthukaruppan Swaminathan^a ; ^b ; ^{muthu@tll.org.sg}Author Vitae ; Pankaj Kumar Yadav^aAuthor Vitae ; Obdulio Piloto^c ; ^{obdulio.piloto@entopsis.com}Author Vitae ; Tobias Sjö ; blom^d ; ^{tobias.sjoblom@igp.uu.se}Author Vitae ; Ian Cheong^a ; ^b ; ^{ian@tll.org.sg}Author Vitae
• 关键词：Poisson-Binomial distribution ; Semi-metric distance ; Non-identical data ; Distance measure ; Image classification ; Image recognition
• 刊名：Pattern Recognition
• 出版年：2017
• 出版时间：March 2017
• 年：2017
• 卷：63
• 期：Complete
• 页码：384-396
• 全文大小：1953 K
• 卷排序：63

文摘

Empirical evidence is provided that real-world data is non-identically distributed. PBR, the first distance measure to account for non-identical data is proposed. PBR was tested in 6 test applications using 12 benchmark data sets. PBR outperforms state-of-the-art measures for most data sets. Avoiding the identical distribution assumption can improve classification.