Logo Recognition with the Use of Deep Convolutional Neural Networks

摘要

Automatic logo recognition is gaining importance dueto the increasing number of its applications. Unlike other objectrecognition tasks, logo recognition is more challenging because ofthe limited amount of the available original data. In this paper,the transfer leaning technique was applied to a DeepConvolutional Neural Network model to guarantee logorecognition using a small computational overhead. The proposedmethod was based on the Densely Connected ConvolutionalNetworks (DenseNet). The experimental results show that for theFlickrLogos-32 logo recognition dataset, our proposed methodperforms comparably with state-of-the-art methods while usingfewer parameters.