COCO-Text: Dataset and Benchmark for Text Detection and Recognition in Natural Images

This paper describes the COCO-Text dataset. In recent years large-scale like SUN and Imagenet drove the advancement of scene understanding and recognition. The goal of COCO-Text is to advance state-of-the-art in detection and recognition in natural images. The dataset is based on the COCO dataset, which contains images of complex everyday scenes. The images not collected with text in mind and thus contain a broad variety of text. To reflect the diversity of text in natural scenes, we annotate text (a) location in terms of a bounding box, (b) fine-grained classification machine printed text and handwritten text, (c) classification into legible illegible text, (d) script of the text and (e) transcriptions of legible. The dataset contains over 173k text annotations in over 63k images. We a statistical analysis of the accuracy of our annotations. In addition, present an analysis of three leading state-of-the-art photo Optical Recognition (OCR) approaches on our dataset. While scene text and recognition enjoys strong advances in recent years, we identify shortcomings motivating future work.