Noise is common phenomenon in every data. Non-text components in binarized text images, which are the results of the text extraction, are considered as noise. It degrades the performance of character recognition module. In this paper, a robust algorithm is proposed to detect text in the binarized text image with noise by extracting a new feature, called stroke width. Firstly, stroke width feature is extracted by finding pairs of stroke boundary pixels. And then text component candidates are detected by employing stroke width feature. Finally, text verification is designed to remove non text components which are mis-detected as text components in the previous step. Experiments on a wide variety of binarized signboard images reveal the feasibility and effectiveness of our proposed approach for detecting text component in noise images.
展开▼