Crystal defect CV

A library to mark defects on images of crystals.

One click usage

Raw Python

Install requiremenst.txt:

pip install -r requirements.txt

Usage:

python main.py [-h] [-k KSIZE] [-t THRESHOLD] [-p POWER] inp out

Arguments:

• [-k KSIZE] (standard=15) -- size of the kernel used. When increased, decreases sensitivity to smaller defects. Can only be odd.
• [-t THRESHOLD] (standard=14.7) -- threshold of the defect intensity to be considered a defect. Basically larger threshold equals smaller marked areas.
• [-p POWER] (standard=1.7) -- basically the same as threshold, but exponential.
• inp -- path to the png image to mark.
• out -- path to save the output png to.

As executable

You can download compiled executable for Windows or Linux here: https://disk.yandex.ru/d/KMenXUanGjCQrw

Using Pyinstaller

Install PyInstaller:

pip install -U pyinstaller

Compile the code into a single-file executable by yourself:

pyinstaller main.py -F

Library usage

First off, import the library:

import crystal_defect_cv as cdcv

Basic usage:

png = cdcv.open_png(DIRECTORY)
defect_matrix = cdcv.find_defects_probs(png)
marked_png = cdcv.mark_defects(png, defect_matrix)
cdcv.save_png(marked_png, SAVE_DIRECTORY)

All library contents:

• open_png(img_path) -- opens the image.
• find_defects_probs(png) -- creates a matrix of where the algorithm thinks the defects are at.
• mark_defects(png, defect_matrix, lighten_up=True) -- marks all nonzero pixels in defect_matrix on the lightened up png.
• save_png(png, save_path) -- saves the png to save_dir.
• modules_dict -- dictionary, that has the processing modules with their name as key. Every processing module is a function similar to find_defects.
• config -- json-based dictionary of configurations. Described in crystal_defect_cv/processing_modules/config.py. Can be changed as a dictionary to configure modules.
• mark_all_in_directory(operator, directory, save_directory) -- uses operator (like modules_dict["sobel_technique"]) on all the png images in the directory and saves marked images in output_directory.