/Parallel-Implementation-of-Edge-Detection-Algorithms

CUDA Implementation of New edge detection algorithms based on adaptive estimation filters

Primary LanguageC

New edge detection algorithms based on adaptive estimation filters

Introduction

This is a parallel CUDA implementation in C of the algorithm proposed in the research paper by Woodhall et al.

Woodhall, M. C., and C. S. Lindquist. "New edge detection algorithms based on adaptive estimation filters." Conference Record of the Thirty-First Asilomar Conference on Signals, Systems and Computers (Cat. No. 97CB36136). Vol. 2. IEEE, 1997.

Understanding the Algorithm

This can be explained step wise :

  1. Padding the image
  2. Computing local mean of each pixel
  3. Computing local variance of each pixel
  4. Computing global variance
  5. Sobel Filter applied for edge detection

Results

Tested on an NVIDIA GeForce GT 640 GPU using the NVIDIA Nsight Eclipse Edition IDE

Original

Screenshot

Sobel Filter Applied (Horizontal)

Screenshot

Notes

  1. serial.py : Serial code in python implemented, code not pythonic in nature. Serves as pseudo code for needs to be done in C.

  2. serial.c : Serial code in C implemented, needs a g++ compiler to run. Output stored in image.png

  3. stb_image.h, stb_image_write.h : public libraries, used for reading and writing images in C.