Digit Classifier

A K-Nearest Neighbours hand written digit classifer written in 32-bit x86 NASM assembly. Utilizes the MNIST hand written digit dataset and euclidean distance to determine its 10 nearest neighbours.

See Releases for compiled versions.

Current accuracy 80% on testing data (Custom inputs will vary)

Important

Usage ./program ./path_to_data_file
Expects a 784 byte file with 28x28 unsigned grey scale pixels
See Utility Scripts to convert an image into expected format

Prerequisites

• 32 Bit Compatible System
• Python 3.0+ for utility scripts
• gcc (version 14.2.1 Recommended)
• NASM (version 2.16.03 Recommended)
• Any Image Editor

Utility Scripts

convertImage.py

Converts any 28x28 image (PNG, JPG) into a grey scale unsigned byte array and writes it to the output file
python convertImage.py ./input.png ./output.data

tester.py

Calculates accuracy based on the first 200 images in the testing file
python tester.py

Build Instructions

Method 1 : Compile with Script

Use the build script.
Make the script executable with chmod +x build.sh
Run the script with ./build.sh

Method 2 : Manual Compilation

Compile manually with nasm and gcc
Compile the assembly
nasm -f elf main.asm
Link with gcc to create the executable
gcc -m32 -o program main.o