jsochacki / BLT

BLT: a Boundary Limited Text extraction tool

BLT: a Boundary Limited Text extraction tool

BLT is a C++ based tool that is aimed at performing automatic list creating for cataloging books, videogames, and other objects that have text regions that are separated with clear and srong boundaries. In order to use BLT one must only take pictures of the items that they want to catalog in their natural state like the following.

Once you have one of these or a whole set of these just place these in the directory you launch BLT from and run BLT. THe result will be held in a .csv file in the same directory. There are similar things for books out there and there are OCR apps but none that I could find that met this need specifically.

Theory or operation:

1. Use Canny edge detection from OpenCV(https://docs.opencv.org/trunk/da/d22/tutorial_py_canny.html) to get the edges
2. Draw bounding boxes around each contour https://docs.opencv.org/3.4/da/d0c/tutorial_bounding_rects_circles.html
3. Snap each enclosed area to it’s own image file
4. Use tesseract-ocr https://github.com/tesseract-ocr/tesseract to get the text from each image
   1. Take an image and use tesseract, run it for all 4 orientations (base, rot 90, rot 180, rot 270) and save the result with the highest score
5. Write it all to a file using standard library calls

Installation

Installation is automatically done with the provided .sh script for ubuntu systems

Building the Project

The project can be built using the provided Makefile. Execute the command below to create the main executable.

make

The Makefile will compile all contents of the src folder into a single executable.

Running the Application

After building the application, the executable will be named main and will be in the root of the build/ folder. Currently the executable can be used to run Canny Edge on all images located inside of a folder. Below is an example execution of the program.

./build/main --threshold1=0 --threshold2=4 --src=./images/ --out=./out/

The source folder needs to exist and contain images files, by default the images/ folder in the root of the project folder will be used. The output folder does not need to exist and will be made if it does not exist. The two thresholds control the two thresholds of the Canny Edge detection. For more information, run the help command below.

./build/main --help

NOTE: Currently OpenCV filesystem tools do not process the ~ character in paths properly. Instead use absolute path in situations where the home directory is required.