本文提出了一种基于深度神经网络的高效自然场景文本检测模型IEAST (Improved EAST).该模型基于EAST (An Efficient and Accurate Scene Text Detector)进行了一系列改进,同时结合了物体检测以及实例分割的思想,通过用Inception Module替换原本特征融合的方式、加入Path Aggregation的思想以及采用Fuzzy Mask来提升模型的整体性能.IEAST是一段式模型,其不仅检测准度上不落后于绝大多数二段式State-of-the-art的文本检测模型,还能保证极快的检测速度.
Abstract
Image Text Detection is the process of accurately locating text area in given images
which is the important prerequisite of OCR (Optical Character Recognition). This paper introduces Improved IEAST (Improved EAST) which is an efficient and accurate scene text detection model based on deep neural network. This model which combines the knowledge of both object detection and instance segmentation is a variant of EAST (An Efficient and Accurate Scene Text Detector) with several improvements. By replacing conv3x3 feature merging with inception module
adding the idea of Path Aggregation and adopting Fuzzy Mask during training
IEAST gains improvements in performance with little compensation. Though IEAST is a one-stage model
it outperforms most of state-of-the-art two-stage text detection models by not only maintaining comparable accuracy but also inferencing faster. IEAST shows good result on ICDAR (International Conference on Document Analysis and Recognition) 2015 dataset
关键词
Keywords
references
MILLETARI F, NAVAB N, AHMADI S A. V-Net: fully convolutional neural networks for volumetric medical image segmentation[C]. In Proc. of 3DV, USA CA: IEEE, 2016: 565-571.
LIAO M H, SHI B G, BAI Xet al. TextBoxes: A fast text detector with a single deep neural network[C]. In AAAI, California: AAAI Press, 2016: 4164-4167.
VENKATESWARLU K, SREERAMA M V. Text detection on scene images using mser[J]. International Journal of Research in Computer And Communication Technology, 2015, 4(7):452-456. DOI: 452
LIU S, QI L, QIN H F et al. Path aggregation network for instance segmentation[C]. In Proc. of CVPR, USA UT: IEEE, 2018: 8759-8768.
ZHOU X Y, YAO C, WEN H et al. EAST: an efficient and accurate scene text detector[C]. In Proc. of CVPR, Honolulu: IEEE, 2017: 2642-2651.
ZHONG Z Y, JIN L W, HUANG S P. DeepText: A new approach for proposal generation and text detection in natural images[C]. In ICASSP, LA: IEEE, 2017: 1208-1212.
ZHONG Z Y, SUN L, HUO Q et al. An anchor-free region proposal network for faster r-cnn based text detection approaches[J]. IJDAR 2019, 22(3):315-327. DOI: 315
LIN T Y, DOLLAR P, GIRSHIK R B et al. feature pyramid networks for object detection[C]. In Proc. Of CVPR, Honolulu: IEEE, 2017: 936-944.
MA J Q, SHAO W Y, YE H et al. Arbitrary-oriented scene text detection via rotation proposals[J]. IEEE Trans. Multimedia, 2017, 20(11):3111-3122. DOI: 3111
NEUMANN L, MATAS J. A method for text localization and recognition in real-world images[C]. In Proc. of ACCV, German: Springer, 2010: 770-783.