Fisher vector for scene character recognition: A comprehensive evaluation
作者：
Highlights：
• We evaluate the Fisher Vector with linear classifier for character recognition.
• Experimental results show that FV outperforms most methods on eight datasets.
• FV could even outperform CNN based methods given limited amount of training samples.
• We show that spatial information is very useful for character representation.
• The results also imply the potential of FV to represent unseen categories.
摘要
•We evaluate the Fisher Vector with linear classifier for character recognition.•Experimental results show that FV outperforms most methods on eight datasets.•FV could even outperform CNN based methods given limited amount of training samples.•We show that spatial information is very useful for character representation.•The results also imply the potential of FV to represent unseen categories.
论文关键词：Character representation,Character recognition,Fisher vector (FV),Gaussian Mixture Models (GMM),Bag of visual words (BOW)
论文评审过程：Received 19 October 2016, Revised 22 May 2017, Accepted 16 June 2017, Available online 23 June 2017, Version of Record 3 July 2017.
论文官网地址：https://doi.org/10.1016/j.patcog.2017.06.022
