Detecting images with text

Introduction

This is a python implementation to recognize images with text in them. It uses OpenCV to detect text within natural scene images using the EAST text detector.

Installation

• Clone this repository and move to the main directory

git clone https://github.com/mmayank74567/detecting_images_with_text.git

cd detecting_images_with_text/

• Create a virtual environment. Install dependencies from the requirements.txt file using the followng command:

pip3 install -r requirements.txt

• EAST text detector requires to have OpenCV 3.4.2 or higher. To install OpenCV, run:

pip install opencv-python

Structure of the directory

The structure of the directory is represented below:

├── detecting_text
│   ├── __init__.py
│   └── detecting_text_in_images.py
├── model
│   └── frozen_east_text_detection.pb
├── output
├── sample
|   ├── test1.jpg
|   ├── test2.jpg
|   ├── test3.jpg
|   ├── test4.jpg
|   ├── test5.jpg
|   ├── test6.jpg
|   └── test7.jpg
└── run.py

• The detecting_text package contains the detecting_text_in_images.py module. The DetectingText class of this module holds two methods, namely, non_max_suppression and east_detection.

EAST Text detector

The east_detection method implements the EAST text detector which is based on the paper EAST: An Efficient and Accurate Scene Text Detector. Since it is an end-to-end deep learning model, the EAST pipeline is highly efficiently as it sidesteps computationally draining sub-algorithms used by other text detectors. The method utilizes the EAST scene text detector model file that is stored in the model folder to detect texts in images.

Non-maxima suppression

The non_max_suppression method ignores the redundant and overlapping bounding boxes by utilizing non-maximum suppression.

• The sample folder contains a mix of natural scene images to demonstrate the implementation. The output folder stores the copy of images from the sample folder that have text in them.

Running

To run the run.py script, execute the following command:

python run.py --model model/frozen_east_text_detection.pb --imagepath sample --output output

It parses three command line arguments

• --model: path to the model file
• --imagepath: path to the folder of images
• --output: path to the folder storing a copy of images with text in them
• --operation: copy to make copies of images with text in them in the output folder or move to move the images with text in them to the output folder (default = copy)

Output

As shown in image below, images with text in them are successfully detected and copied to the output folder.