Scalable Computing Assignment

Edge detection by applying Stencil code on Image

In this assignment we will use a Stencil code. An Image is considered to be a matrix of pixels. And stencil operators take an input image and convolve it with Stencil kernel to produce the output image according to the following formula:

We will use at nine point stencil that performs edge detection. Qx,y is a pixel of the output image, which lies at the intersection of the row x and the column y. Px,y is a pixel of the input image, which lies at the intersection of the row x and the column y. Indices of P and Q shows the position of the pixels in the input image and the output image, respectively.

Cij also shows the elements of the 3*3 stencil matrix showed in the following figure:

In the basic implementation, we could keep our input and output images as arrays of floating point numbers. To apply the Stencil, we may set up two arrays that traverse all pixels in the image and for each pixel we take the pixel and its eight neighbors, multiply them by the weights in the Stencil matrix, add them, and this produces the output image value.

We have to then check whether this value is within the bounds of our intensity depth. Specifically this brightness has to be between 0 and 255 for an 8bit grey scale image.

The task is to implement an initial serial version of the program, where it takes an image as an input and then produces an output image after applying the stencil matrix on the input image. The code is optimized the in two levels: · Firstly, using openMP a multithreaded version of the program is created.

· Secondly, the code is further optimizes using MPI in order to distribute the computations to different compute nodes in the cluster in the Linux Pool.