Abstract
Scene text spotting is currently a popular research topic in the computer vision community. However, it is a challenging task due to the variations of texts and clutter backgrounds. In some situation, especially in the natural scene, it is not necessary to understand the whole text information in an image if we have acquired the key information. In this paper, we propose a bottom-to-up framework for keywords spotting and context extraction for multi-oriented Chinese in scene images. The proposed framework includes three modules which are Chinese character detection, keywords spotting and context extractor. We adopt the object detection methods to detect and recognize the Chinese text simultaneously in character-level. The geometric relationship between keywords and their context is analyzed, which is robust to multi-oriented text. The proposed framework classifies the valuable images, extracts possible keywords context and achieves a good performance in the CTW dataset.