HDSRNet: A Simple Segmentation-free Method for Unconstrained Handwritten Digit String Recognition

Handwritten digit string recognition (HDSR) is one of the most challenging tasks in the area of off-line handwritten optical character recognition. The main challenge comes from the segmentation errors caused by several factors such as touching, breaking, complex background, and unknown length of string. In this paper, we propose a new segmentation-free HDSR system using a one-stage object detector based on a convolutional neural network (CNN). This is a simple, efficient, and lightweight algorithm. This work describes the following: (1) the problem of HDSR can be solved successfully by CNN-based object detectors which determine both locations and classifications for objects and it makes us completely free from the complicated segmentation; (2) the network for HDSR does not need to be too heavy and its depth can be determined suitably by the size of receptive field; (3) it is efficient to use single-scale and single-anchor method rather than multi-scale and multi-anchor one, in the detector for HDSR. The experiments conducted on the benchmark datasets (CVL, ORAND-CAR) show that the proposed method achieves excellent performance compared to other state-of-the-art methods.