This project develops a character recognition system using a Random Forest Classifier, it classifies characters from various images, crucial for applications such as automatic license plate recognition. Below is an example of an image from the dataset used in this project:
| Filename | Identified License Plate |
|---|---|
| CIN20356.jpg | CIN20356 |
| CMG21FG.jpg | CMG21FG |
| FSD23429.jpg | FSD23429 |
| PCT15PY.jpg | PCT15PY |
| PGN141GR.jpg | PGN141GR |
| PGN756EC.jpg | PGN756EC |
| PKL8C63.jpg | PKL8C63 |
| PKRR788.jpg | PKRR788L |
| PKS30W3.jpg | PKS30W3 |
| PO033AX.jpg | POO33AX |
| PO096NT.jpg | PO096NT |
| PO155KU.jpg | PO155KU |
| PO2J735.jpg | PO2J735 |
| PO2W494.jpg | PO2W494 |
| PO522WJ.jpg | PO522WJ |
| PO5T224.jpg | PO5T224 |
| PO6K534.jpg | PO6K534 |
| PO778SS.jpg | PO778SS |
| POBTC81.jpg | POBTC81 |
| POZS221.jpg | POZS221 |
| PSE22800.jpg | PSE22800 |
| PSZ47620.jpg | PSZ47620 |
| PZ0460J.jpg | PZ0460 |
| PZ492AK.jpg | PZ492AK |
| WSCUP62.jpg | WSC7UG62 |
| ZSL17729.jpg | ZSL17729 |
- Image Processing: The images are initially processed to improve the classifier's accuracy. This involves loading images from a dataset, converting them to grayscale to reduce complexity, applying a binary threshold to separate characters from the background, and dilating to emphasize features.
- Data Preparation: Each processed image is resized to a uniform dimension (120x120 pixels) and flattened into a one-dimensional vector. This vectorization transforms the image data into a format suitable for machine learning models.
- Model Training: Utilizing the RandomForestClassifier for training on processed image data.
- Model Evaluation: After training, the model's effectiveness is evaluated through metrics such as precision, recall, and F1-score across a test dataset. These metrics help quantify the model’s ability to classify each character correctly.
- Model Serialization: The fully trained model is serialized using Python’s
picklemodule, enabling it to be saved and reloaded for future use without retraining.
characters_dataset/: Contains labeled images of characters. Each label directory houses images representing a specific character, aiding in supervised learning.characters_recognizer_rf.pkl: This file is a serialized version of the trained Random Forest model, ready for deployment in character recognition tasks.templates/: directory contains a comprehensive template featuring all the characters used on license plates. This template is designed with a font style closely resembling that found on vehicle registration plates, making it ideal for training the character recognition system.
- Image Loading and Processing: This crucial step involves several image manipulations to prepare data for training:
- Grayscale Conversion: Converts the RGB image to grayscale to reduce computational load.
- Thresholding: Applies a binary threshold to make the image binary, which is useful for isolating characters from backgrounds.
- Dilation: Enhances features of the character in the binary image.
- Resizing and Flattening: Standardizes the size of images and flattens them into arrays for machine learning processing.
- Dataset Compilation: Compiles the feature vectors (
X) and labels (y) from the processed images, ensuring accurate correspondence between features and targets.
- Training Data Split: The dataset is split into 80% for training and 20% for testing, maintaining a balance between learning and validation capabilities to prevent overfitting.
- Random Forest Classifier: Trains on the flattened image data, learning to recognize and classify each character based on its features.
- Performance Evaluation: The model’s performance is evaluated using classification metrics such as precision, recall, and F1-score, providing insights into its accuracy and effectiveness.
| precision | recall | f1-score | support | |
|---|---|---|---|---|
| 0 | 1.00 | 1.00 | 1.00 | 38 |
| 1 | 1.00 | 1.00 | 1.00 | 31 |
| 2 | 1.00 | 1.00 | 1.00 | 29 |
| 3 | 1.00 | 1.00 | 1.00 | 35 |
| 4 | 1.00 | 1.00 | 1.00 | 35 |
| 5 | 1.00 | 1.00 | 1.00 | 33 |
| 6 | 1.00 | 1.00 | 1.00 | 29 |
| 7 | 1.00 | 1.00 | 1.00 | 30 |
| 8 | 1.00 | 1.00 | 1.00 | 29 |
| 9 | 1.00 | 1.00 | 1.00 | 36 |
| A | 1.00 | 1.00 | 1.00 | 36 |
| B | 1.00 | 1.00 | 1.00 | 34 |
| C | 1.00 | 1.00 | 1.00 | 26 |
| D | 1.00 | 1.00 | 1.00 | 22 |
| E | 1.00 | 1.00 | 1.00 | 25 |
| F | 1.00 | 1.00 | 1.00 | 34 |
| G | 1.00 | 1.00 | 1.00 | 28 |
| H | 1.00 | 1.00 | 1.00 | 32 |
| I | 1.00 | 1.00 | 1.00 | 23 |
| J | 1.00 | 1.00 | 1.00 | 26 |
| K | 1.00 | 1.00 | 1.00 | 27 |
| L | 1.00 | 1.00 | 1.00 | 25 |
| M | 1.00 | 1.00 | 1.00 | 25 |
| N | 1.00 | 1.00 | 1.00 | 31 |
| O | 1.00 | 1.00 | 1.00 | 31 |
| P | 1.00 | 1.00 | 1.00 | 25 |
| R | 1.00 | 1.00 | 1.00 | 29 |
| S | 1.00 | 1.00 | 1.00 | 30 |
| T | 1.00 | 1.00 | 1.00 | 42 |
| U | 1.00 | 1.00 | 1.00 | 30 |
| V | 1.00 | 1.00 | 1.00 | 38 |
| W | 1.00 | 1.00 | 1.00 | 30 |
| X | 1.00 | 1.00 | 1.00 | 31 |
| Y | 1.00 | 1.00 | 1.00 | 25 |
| Z | 1.00 | 1.00 | 1.00 | 28 |
| precision | recall | f1-score | support | |
|---|---|---|---|---|
| accuracy | 1.00 | 1058 | ||
| macro avg | 1.00 | 1.00 | 1.00 | 1058 |
| weighted avg | 1.00 | 1.00 | 1.00 | 1058 |