Development of a Low-cost Industrial OCR System with an End-to-end Deep Learning Technology
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Optical character recognition (OCR) has been studied for decades because it is veryuseful in a variety of places. Nowadays, OCR's performance has improved significantly due tooutstanding deep learning technology. Thus, there is an increasing demand for commercial-gradebut affordable OCR systems. We have developed a low-cost, high-performance OCR system forthe industry with the cheapest embedded developer kit that supports GPU acceleration. Toachieve high accuracy for industrial use on limited computing resources, we chose astate-of-the-art text recognition algorithm that uses an end-to-end deep learning network as abaseline model. The model was then improved by replacing the feature extraction network withthe best one suited to our conditions. Among the various candidate networks, EfficientNet-B3has shown the best performance: excellent recognition accuracy with relatively low memoryconsumption. Besides, we have optimized the model written in TensorFlow's Python API usingTensorFlow-TensorRT integration and TensorFlow's C++ API, respectively.
Abstract
Optical character recognition (OCR) has been studied for decades because it is veryuseful in a variety of places. Nowadays, OCR's performance has improved significantly due tooutstanding deep learning technology. Thus, there is an increasing demand for commercial-gradebut affordable OCR systems. We have developed a low-cost, high-performance OCR system forthe industry with the cheapest embedded developer kit that supports GPU acceleration. Toachieve high accuracy for industrial use on limited computing resources, we chose astate-of-the-art text recognition algorithm that uses an end-to-end deep learning network as abaseline model. The model was then improved by replacing the feature extraction network withthe best one suited to our conditions. Among the various candidate networks, EfficientNet-B3has shown the best performance: excellent recognition accuracy with relatively low memoryconsumption. Besides, we have optimized the model written in TensorFlow's Python API usingTensorFlow-TensorRT integration and TensorFlow's C++ API, respectively.
참고문헌 (26)
- M. Jaderberg, "Synthetic Data and Artificial Neural Networks for Natural Scene Text Recognition", pp.1~10 (2014)
- A. Poznanski, "CNN-N-gram for Handwriting Word Recognition", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp, pp.2305~2314 (2016)
- Shangbang Long, "Scene Text Detection and Recognition: The Deep Learning Era", pp.1~20 (2018)
- M. Jaderberg, "Reading Text in the Wild with Convolutional Neural Networks", International Journal of Computer Vision, 116(1), pp.1~20 (2016)
- M. Liao, "Textboxes: A Fast Text Detector with a Single Deep Neural Network", Proceedings of Advancement of Artificial Intelligence, pp.4161~4167 (2017)
- Z. Tian, "Detecting Text in Natural Image with Connectionist Text Proposal Network", Proceedings of European Conference on Computer Vision, pp.56~72 (2016)
- X. Zhou, "East : An Efficient and Accurate Scene Text Detector", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp.5551~5560 (2017)
- Christian Bartz, "STN-OCR: A Single Neural Network for Text Detection and Text Recognition", pp.1~9 (2017)
- M. Busta, "Deep Textspotter : An End-to-end Trainable Scene Text Localization and Recognition Framework", Proceedings of IEEE International Conference on Computer Vision, pp.2204~2212 (2017)
- T. He, "An End-to-end Textspotter with Explicit Alignment and Attention", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp.5020~5029 (2018)
- X. Liu, "Fots : Fast Oriented Text Spotting with a Unified Network", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp.5676~5685 (2018)
- "Comparison of Deep-learning Software"[URL]
- M. Liao, "TextBoxes++ : A Single-shot Oriented Scene Text Detector", Journal of IEEE Transactions on Image Process, 27(8), pp.3676~3690 (2018)
- B. Shi, "An End-to-end Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition", Journal of IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(11), pp.2298~2304 (2017)
- K. Simonyanm, "Very Deep Convolutional Networks for Large-Scale Image Recognition", Proceedings of International Conference on Learning Representations, pp.7~9 (2015)
- K. He, "Deep Residual Learning for Image Recognition", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp.770~778 (2016)
- K. Kim, "PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection", pp.1~7 (2016)
- A. G. Howard, "MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications", pp.1~9 (2017)
- M. Sandler, "MobileNetV2 : Inverted Residuals and Linear Bottlenecks", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp.4510~4520 (2018)
- A. Howard, "Searching for MobileNetV3", Proceedings of IEEE International Conference on Computer Vision, pp.1314~1324 (2019)
- M. Tan, "EfficientNet: Rethinking Model Scaling for Convolution Neural Networks", pp.1~10 (2019)
- Sandler, "Mobilenetv2 : Inverted Residuals and Linear Bottlenecks", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp.4510~4520 (2018)
- Tan, "MnasNet : Platform-aware Neural Architecture Search for Mobile", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp.2820~2828 (2019)
- Hu, "Squeeze-and-excitation Networks", Proceedings of IEEE Conferenc on Computer Vision and Pattern Recognition, pp.7132~7141 (2018)
- "Multi-dimensional Arrays with Broadcasting and Lazy Computing"[URL]
- "NumCpp"[URL]