ju-chang / vivado-stream-convolution

README

This is source code repository for my CSE_291: FPGA Image Processing Course Homework 2.

PID: A53095838

Link to the Video

https://www.youtube.com/watch?v=S7s1nkXVrHc

How to run the Project

## in the vivada_hls shell
git clone https://bitbucket.org/xyz0/cse_291_fpga_cv_hw2/ hw2  && cd hw2/sobel
vivado_hls ./script.tcl

Architecture

The program uses line buffer to receive streaming image pixel and sliding window buffer to multiply with the kernel. In order to make sure window buffer is filled properly at the beginning, the loop starts with i = 0, j = 0 and output_image[i-1][j-1], so the invalid computing result will be discarded and there is no if/else inside the loop to slow the program down.

Result Comparision

Old

Optimized

Test Bench

I compare every pixel value between the original sobel implementation and this optimized one to determine its correctness and accuracy.

//stuff
sobel(input_image, output_image);
sobel_truth(input_image, output_image_gold);
bool success = true;
for (i=1; i<359; i++)
   for (j=1; j<639; j++)
       if(!output_image[i][j] == output_image_gold[i][j]){
	    printf("Test failed at (%d,%d), truth = %d, yours = %d\n",i,j
			,output_image_gold[i][j],output_image[i][j]);
	    success = false;
	}