CUDA-Accelerated Canny Edge Detection

This project implements the Canny edge detection algorithm using CUDA for GPU acceleration and provides a comparison with a CPU implementation.

---

Project Overview

The primary goal is to perform edge detection on PGM images using the Canny algorithm. It includes the following steps:

• Gaussian smoothing
• Gradient calculation
• Non-maximal suppression
• Hysteresis thresholding

GPU-accelerated (CUDA) and CPU versions are included for performance comparison.

---

File Breakdown

• canny_edge.cu
  Main program logic:

  • Parses command-line arguments (input image, sigma, thresholds)
  • Implements CPU (gaussian_smooth_cpu, derrivative_x_y_cpu) and GPU (gaussian_smooth_gpu, derrivative_x_y_gpu) steps using CUDA kernels
  • Calls functions for magnitude, non-max suppression, and hysteresis
  • Handles image I/O via pgm_io.cu

• hysteresis.cu
  Implements:

  • non_max_supp
  • apply_hysteresis
  • follow_edges

• pgm_io.cu

  • PGM I/O utilities: read_pgm_image, write_pgm_image
  • Some PPM utilities also included but unused in main logic

• Makefile

  • Builds the project with nvcc
  • Contains targets for running and cleaning

• pics/

  • Sample PGM images for testing

---

Build Instructions

Make sure you have CUDA installed.

make

Run: Execute the compiled program

make run

Alternatively :

./canny <image_path.pgm> <sigma> <tlow> <thigh>