Review on text detection methods on scene images

Recently, a lot of effort has been put into developing text detection methods on natural scene images in academic research and industry. In general, text detection refers to localizing all text instances in an image which can be further processed with an Optical Character Recognition (OCR) software in order to obtain machine-readable characters. The amount of published methods is constantly growing which makes it very challenging to be up-to-date with all approaches and state-of-the-art methods. Review papers become outdated in a less than a year from being published. Deep learning, a fast-growing field by itself, has become a mainstream approach in developing text detection methods. In this paper we present the up-to-date state-of-the-art methods in this challenging field. The methods are compared by their accuracy and real-time performance. We also present the most popular evaluation datasets for scene text detection.