/euro-coin-classifier

Primary LanguagePythonMIT LicenseMIT

Euro Coin Classifier

This Euro Coin Classifier is part of the Euro Coin Detector project.
The Main GitHub Repository: Euro Coin Detector

Introduction

This Euro Coin Classifier is part of the Euro Coin Detector project that aims to locate and recognize euro coins from natural images and classify them according to their coin denomination and tell their values.

The training of this classifier uses statistics analytics to extract and process information from many images of the euro coin series of each denomination. The goal is to predict and generalize each euro coin type's attributes using Machine Learning for Predictive Data Analytics techniques.

The classifier is developed using Artificial Intelligence to describe the euro coins' shape, size, color, patterns, etc. so later the classifier can be used to determine whether an arbitrary object is a certain denomination of euro coin.

Data Exploration and Analytics

The reports are focused on Descriptive Statistics analytics. For each euro coin denomination, its HSI (Hue, Saturation, Intensity) and LUV (Luma, Blue Difference, Red Difference) color space values are the main area of interests.

The reports can be found under the reports folder.

Data Preparation

The dataset_collector.py script is used to prepare dataset for the training of the classifier. It uses computer vision techniques to recognize and segment euro coins from natural images, and output the segmented coins into the output folder.

The dataset collector takes natural images of euro coins as input, and process them into segmented image dataset that is ready for the classifier training process.

Usage:

python dataset_collector.py [image_files...]

Multiple image files can be all passed at once to batch process them all.

Example:

Original Processed
Original Processed

Original: A natural image of 22 euro coins scattered on a surface.
Processed: 22 separate images of euro coins cropped to just the coins themselves.

Train Your Own Classifier

After the dataset is ready and placed in the data folder, the training can start by running the classifier_trainer.py script. This script focuses on Descriptive Statistics analytics and process the data and output the reports under the reports folder in .csv format.

Usage:

python classifier_trainer.py

Running the script will process all images in the data folder.

Training

Here's an example report from 1794 samples of 1-euro coins:

Count Min 1st Quart Mean Median 3rd Quart Max Std Dev
Hue 1794 3 10.0 23.17 13.0 15.0 156 30.28
Saturation 1794 3 53.0 98.21 97.0 135.0 228 53.09
Lightness 1794 52 94.0 118.85 115.0 141.0 218 31.94
Luma 1794 48 78.0 102.55 98.0 123.0 209 31.75
Blue Difference 1794 93 112.0 117.80 118.0 124.0 151 7.54
Red Difference 1794 77 134.0 141.26 142.0 149.0 164 9.28

The histograms of the 1-euro coins:

Hue Saturation Intensity
Hue Saturation Intensity

After the training, a classifier will be generated as a JSON file euro_coin_detector_classifier.json.

Licensing

Please see the file named LICENSE.md.

Author

  • Chen Yumin

Contact