基于实例分割方法的场景文本检测算法研究

Research on Text Detection Algorithm of Scene Based on Instance Segmentation Method

本研究主要使用实例分割技术对自然场景中的文字进行检测,以PSENet网络为研究对象,将ResNext101与FPN结合作为主干网络进行特征提取,以提高文本检测模块的精度。本文着重研究基于像素级别的实例分割问题,对文本图像中的文字进行精确定位,实现高效简洁地提取文本图像中的文字。实验结果表明,改进后的模型在进行文本检测上有了更好的效果。

Text can be seen anywhere and everywhere in our daily life, and text is an essential way to convey information in our process. Text detection in natural scenes has been applied in many scenarios, such as the detection and recognition of license plates, and is also used in the field of autonomous driving. At present, in the field of deep learning, the technology for simple text detection in electronic documents is relatively mature, but there are still many difficulties in the detection of text in natural scenes,this project mainly uses the instance segmentation technology to detect text in natural scenes, taking PSENet network as the research object, combining ResNext101 and FPN as the backbone network for feature extraction, which improves the accuracy of the text detection module. We focus on the problem of instance segmentation based on pixel level to achieve efficient and concise extraction of text in text images by pinpointing the text in the text images. It is demonstrated experimentally that the improved model has better results in performing text detection.