Chinese font recognition (CFR) has gained significant attention in recentyears. However, due to the sparsity of labeled font samples and the structuralcomplexity of Chinese characters, CFR is still a challenging task. In thispaper, a DropRegion method is proposed to generate a large number of stochasticvariant font samples whose local regions are selectively disrupted and aninception font network (IFN) with two additional convolutional neural network(CNN) structure elements, i.e., a cascaded cross-channel parametric pooling(CCCP) and global average pooling, is designed. Because the distribution ofstrokes in a font image is non-stationary, an elastic meshing technique thatadaptively constructs a set of local regions with equalized information isdeveloped. Thus, DropRegion is seamlessly embedded in the IFN, which enablesend-to-end training; the proposed DropRegion-IFN can be used for highperformance CFR. Experimental results have confirmed the effectiveness of ournew approach for CFR.
展开▼
