Text Image Enhancement in Scenery Images for Degraded Character Recognition using DCT

Abstract

This paper proposes a method to enhance text images for a system which extracts, recognizes and translates multilingual characters in scenery images captured by a digital camera. The proposed method magnifies text images in frequency domain. It restores high-frequency components by a DCT(Discrete Cosine Transform) based approach with an estimated enlarged image, and reduces mosquito and block noises caused by JPEG(Joint Photographic Experts Group) compression in the enhanced process. The obtained enhanced text images are binarized and then recognized by a commercial character recognition software. Experiments are performed for printed documents, sign boards and plates captured by a digital camera. In our experiments, texts are manually extracted from images because our goal in this paper is image enhancement. Compared with traditional approaches, the recognition ratio for our enhanced images by using a commercial character recognition software improves.