Kazakh Letter Recognition System

Project Date: December, 2023

1. Problem and Motivation

The project, "Kazakh Letters Recognition with CNN", tackles the challenge of recognizing handwritten Kazakh letters, a problem unexplored due to the absence of suitable datasets. Existing technologies lack the capability to process these characters, requiring a custom approach to address this gap. Recognizing handwritten Kazakh letters has significant implications for advancing Kazakh-language technology and improving accessibility.

2. Dataset

To overcome the lack of available datasets, we created our own. We collected data from 144 participants, who contributed a total of 1,296 handwritten characters over 6 hours. The dataset underwent preprocessing, where the images were converted to grayscale, segmented into 150x150 regions, and divided into training, validation, and test sets.

3. Model

A 2D Convolutional Neural Network (CNN) was developed to identify patterns in the handwritten letters. The model architecture consisted of convolutional layers for feature extraction, max-pooling layers for dimensionality reduction, and a final softmax-activated dense layer for classification into 9 distinct classes.

The model was trained on the custom dataset and evaluated using a confusion matrix to measure classification accuracy across all classes. Further individual image testing demonstrated the model's robustness and capability to generalize to new inputs. This comprehensive evaluation confirmed the model's effectiveness in recognizing Kazakh handwritten letters.

Technologies Used

• TensorFlow/Keras
• OpenCV for YOLO
• Custom Dataset

Project Gallery