EasyDeepOCR
Deep Learning Based Optical Character Recognition Library
- Tool designed for industrial markings, like serial numbers, expiry dates, part numbers…
- Very easy to use, simple API and no training required
- Optional topology definition to filter out non-relevant texts
- Focus on text reading capability without compromising speed
- Improve the detection and recognition of the text with dedicated trainings
- Runs on CPU and GPU, Intel and ARM platforms
Description
EasyDeepOCR is a new generation OCR library that relies on deep learning technologies to tackle today’s challenges in text reading. The detection of the texts in the image and the recognition of the characters are fully automated. With the help of a text topology, character ambiguities can be reduced and non-relevant text can be filtered out. An optimization process, which only needs a few samples, can speed up the processing. EasyDeepOCR is a major improvement over EasyOCR and EasyOCR2 libraries, combining ease of use and enhanced capabilities.
Simple API, Yet Impressive Results
With a single “Read” call, all the texts in the image will be detected and read. No more need to setup complex configuration.
Fine-tune text detection and recognition with retraining
Users can retrain the neural network behind EasyDeepOCR’s text detection and recognition. A dedicated project type in Deep Learning Studio provides an efficient interface for annotating the texts and running the training. With as few as 5 images, you can achieve a noticeable improvement on difficult use cases.
Fully Featured Open eVision Studio Tool
The EasyDeepOCR tool in Open eVision Studio exposes all the features of this library. The user can choose the topology and annotate some sample images to perform a pipeline optimization. Deep learning execution settings are exposed, to choose between different Devices (CPU, GPU) and Engines (OpenVINO, TensorRT…).
Topology Definition To Focus On The Right Texts
Topology defines the format of the targeted text. Topology is very practical in the context of industrial marking, to focus on expiry date, part number or other strongly formatted texts. The definition of Topology allows the user to get only the required texts but also helps the character recognition by reducing ambiguities.
Good Performances On CPU (Intel OpenVINO) And GPU (NVidia TensorRT)
EasyDeepOCR has been optimized to run on Intel and ARM CPU (thanks to OpenVINO) and NVidia GPU (thanks to TensorRT).
Open eVision Studio
Complex image processing sequences can be designed using a graphical interface. The collection of tools represents the diversity and the capabilities of the Open eVision libraries. The C++, Python and C# source code, corresponding to the processing pipeline, is automatically generated and provides an interactive documentation of the Open eVision API. Open eVision Studio can process live image sources such as a GigE Vision camera, a Coaxlink frame grabber or eGrabber recorder sequences.
This application is free of charge, runs on Windows, Linux and is compatible with Intel and ARM 64-bit architectures.
