This paper proposes a region based convolutional neural network forsegmentation-free word spotting. Our net- work takes as input an image and aset of word candidate bound- ing boxes and embeds all bounding boxes into anembedding space, where word spotting can be casted as a simple nearestneighbour search between the query representation and each of the candidatebounding boxes. We make use of PHOC embedding as it has previously achievedsignificant success in segmentation- based word spotting. Word candidates aregenerated using a simple procedure based on grouping connected components usingsome spatial constraints. Experiments show that R-PHOC which operates on imagesdirectly can improve the current state-of- the-art in the standard GW datasetand performs as good as PHOCNET in some cases designed for segmentation basedword spotting.
展开▼
