Writer Identification and Writer Retrieval Based on NetVLAD with Re-ranking

  • 2021-02-22 18:27:50
  • Shervin Rasoulzadeh, Bagher Babaali
  • 0

Abstract

This paper addresses writer identification and writer retrieval which isconsidered as a challenging problem in the document analysis and recognitionfield. In this work, a novel pipeline is proposed for the problem at hand byemploying a unified neural network architecture consisting of the ResNet-20 asa feature extractor and an integrated NetVLAD layer, inspired by the vector oflocally aggregated descriptors (VLAD), in the head of the latter part. Havingdefined this architecture, the triplet semi-hard loss function is used todirectly learn an embedding for individual input image patches. Subsequently,generalized max-pooling technique is employed for the aggregation of embeddeddescriptors of each handwritten image. Also, a novel re-ranking strategy isintroduced for the task of identification and retrieval based on $k$-reciprocalnearest neighbors, and it is shown that the pipeline can benefit tremendouslyfrom this step. Experimental evaluation has been done on the three publiclyavailable datasets: the ICDAR 2013, CVL, and KHATT datasets. Results indicatethat while we perform comparably to the state-of-the-art on the KHATT, ourwriter identification and writer retrieval pipeline achieves superiorperformance on the ICDAR 2013 and CVL datasets in terms of mAP.

 

Quick Read (beta)

loading the full paper ...