MenuParser is an restaurant menu parser using Tesseract and OpenCV written in Python 3. It takes in a picture or a photo of a restaurant menu, and spits out a text file and a JSON file with the item name and price.
This is achieved by utilizing Tesseract Optical Character Recognition technology, along with OpenCV's image manipulation.
Unfortunately, it can only be used on a small chunk of the menu at a time. It also can't be used on poor quality pictures or photos, because of the OCR technology limitations.
- First, you need to make sure the image is parsable. A region of the menu must be chunked out and rows should be distinct.
For example, you have to chunk the region like this:
Something Delicious 1.99
Something Really Delicious 13.99
Coke or something similar 1.00
It does not have to have aligned columns.
What not to do:Something Delicious 1.99 Something Really Delicious 13.99 - Second, make sure to have your image file within the same directory as the main folder, and run like this.
MenuParser.py [image_file] --output [destination_file]
This runs the raw picture without any image manipulation. See if this works out well. If not, proceed.
There are arguments which can help with the OCR accuracy: denoise and greyscale.
Denoise reduces the grainy pixels on digitally degraded images. These griany pixels confuse the OCR algorithm, and causes inaccuracy.
Greyscale changes the image to black and white, thus easier to recognize text.
These features don't always grant better results. Please use best judgement after playing around with the settings.
Denoise:-d, --denoiseie:MenuParser.py [image_file] --output [destination_file.txt] -d
Greyscale:-g, --greyscaleie:MenuParser.py [image_file] --output [destination_file] -g - Third, you will have two files in
outputfolder:[destination_file.txt]and[destination_file.txt]_parsed.JSON
The JSON file is very easy to work with, but do check the .txt file as well for more accuracy.
1. Here is the initial test menu chunk. It only works if the chunk is like this:
2. Run the program
3. The initial text file from Optical Character Recognition
4. JSON parsing done through the program
As you can see, the OCR is far from perfect. You will have to do some tweaking before copying and pasting. However, this is far better than having to manually input letter by letter.
- Python 3.6.5 or higher
- Tesseract 3.x or higher for Linux, Tesseract 5.0 for Windows
- pytesseract 0.3.4
- opencv-python 4.2.0.34
07/03/2020 - Added Exceptions