Detecting Text In Natural Scene Based On Random Width Histogram

With the fast development of the smart phone on the mobile market, the product based on computer vision technology has been widely applied in recent years, which includes face detection, gesture recognition and other intelligent application. All these fantastic programs are encouraged by various detection algorithms. As an important part of the field of object detection, text detection has been received more and more attention. Among different text detection algorithms, the research on text detection in natural scene has been a popular topic in academia. Distinct from the text content in the scanned documents, text in natural scenes is difficult to detect due to large variation in shapes, textures, and colors. Therefore, the research on text detection in natural scenes has great significance, especially for improving detection accuracy and the property of robustness.Considering the complexity of natural scene, it is necessary to create more robust feature to represent text and to distinguish text from non-text area. Due to the strong stability and consistency of the width of text, this paper focuses on the text detection algorithm based on stroke width transform (SWT). By analysis of the SWT algorithm, we found that pixels near the corners of a stroke did not indicate the expected stroke width, and we believed that this was a result of how strokes were defined based on paralleled edges. In order to overcome these shortcomings, this paper proposes a novel text detector, random width histogram(RWH). With the help of local binary pattern (LBP) and support vector machine (SVM) algorithm, we achieve a state-of-art text detection algorithm.In order to provide a baseline comparison, we ran our algorithm on the publicly available datasets, ICDAR dataset and Street Vision Text dataset. The experimental results show that the performance of our proposed algorithm outperforms all results from other algorithms.