deeptext

A cross-platform framework for deep learning based text detection, recoginition and parsing

Getting started

Installation

• Install using conda for Linux, Mac and Windows (preferred):

conda install -c fcakyon deeptext

• Install using pip for Linux and Mac:

pip install deeptext

Install teserract-ocr for text recognition.

Basic Usage

# import package
import deeptext

# set image path and export folder directory
image_path = 'idcard.png'
output_dir = 'outputs/'

# apply text detection and export detected regions to output directory
detection_result = deeptext.detect_text(image_path, output_dir)

# apply text recognition to detected texts
recognition_result = deeptext.recognize_text(image_path=detection_result["text_crop_paths"])

Advanced Usage

You can pass filter parameters if you want to scrap texts from image by predefined regions.

# import package
import deeptext

# set image path and export folder directory
image_path = 'idcard.png'
output_dir = 'outputs/'

# define regions that you want to scrap, by quad (box) points
filter_params = {"type": "box"
                 "boxes": [[[0.1460 , 0.0395],
                            [0.8417, 0.0535],
                            [0.8412, 0.1099],
                            [0.1455, 0.0959]],
                           [[0.3467, 0.3398],
                            [0.5417, 0.3535],
                            [0.5412, 0.4099],
                            [0.3455, 0.3959]]],
                 "marigin_x": 0.05,
                 "marigin_y": 0.05,
                 "min_intersection_ratio": 0.9}

# or define regions that you want to scrap, by centroids
filter_params = {"type":"centroid",
                 "centers": [[0.44, 0.49],[0.49, 0.08]],
                 "marigin_x": 0.03,
                 "marigin_y": 0.05}

# apply craft text detection in predefined regions and export detected regions to output directory
detection_result = deeptext.detect_text(image_path,
                                         output_dir,
                                         detector="craft",
                                         filter_params=filter_params)

# apply tesseract (eng) text recognition to detected texts
recognition_result = deeptext.recognize_text(image_path=detection_result["text_crop_paths"],
                                             recognizer="tesseract-eng")

Updates

6 April, 2020: Conda package release

3 April, 2020: Tesseract text recoginition and positional text scraping support

30 March, 2020: Craft text detector support

TODO

• Craft text detection (inference)
• Ctpn text detection (inference)
• Psenet text detection (inference)
• Tesseract text recoginition (inference)
• Aster text recognition (training and inference)
• Moran text recognition (training and inference)
• Positional text scraping