image-processing

Parallel Edge Detection using the Sobel Operator

The main idea behind the Sobel Operator is to find the gradient (rate and direction of change in pixel brightness). High gradient: usually indicates an edge, as it means a sharp change in brightness

---

Logic:

• We can represent each pixel in a grey-scale image as a 3x3 matrix where each pixel is the middle element and the 8 surrounding pixels are the 8 surrounding elements
• To process an image properly (find the edges), we will use 2 matrices, one for vertical processing and the other for horizontal. these metrices are known as the Sobel Kernels and they highlight changes in brightness
• Horizontal matrix Gx is [[-1, 0, 1], [-2,0,2], [-1,0,1]] while vertical matrix Gy is [[-1,-2,-1],[0,0,0],[1,2,1]]
• If we multiply each pixel and its neighbors by each of Gx and Gy, and then summing each of the outputs, we can get 2 numbers representing the change intensity horizontally and vertically
• By calculating the Euclidean distance "G" (the root of the sum of Gx and Gy each squared), we can get the "average" of the vertical and horizontal processing results
• The resulting vertical and horizontal intensity will be compared to T (threshold) - since each pixel can be from 0-255, the threshold of change can be 256/2 or 128
• If G > T: It is considered an edge pixel

---

Compile and run main code

• cd src
• g++ -O2 -fopenmp -o main main.cpp parallel.cpp sequential.cpp cacheTest.cpp $(pkg-config --cflags --libs opencv4) ./main

How to run MPI code

• cd src
• mpicxx -std=c++17 mpi.cpp -o mpi_sobel pkg-config --cflags --libs opencv4
• mpirun -np 4 ./mpi

Get cache test data

perf stat -e cache-misses,cache-references ./main