- objective: create a canny edge detector in C, run on test images, and compare to expected output.
- chess picture images: http:https://www.cs.ucf.edu/courses/cap4453/inputpics
- canny images: http:https://www.cs.ucf.edu/courses/cap4453/outputpics/
- canny algorithm is same as sobel, except for the "smoothener"
- canny smoothener: gaussian
- sobel smoothener: four one's
- instructions to test whether part 1 works:
- set sigma value to 1.0
- run code on garb34.pgm
- check: if output image is similar to the magnitude image produced by sobel.c, code is correct
- check: output should look like cannymag.pgm### part 2:
- perform "max test" by moving through magnitude vector array and finding gradient magnitude and direction
- pseudocode:
/*
current pixel: [i, j]
[i-1, j-1] [i-1, j] [i-1, j+1]
[i, j-1] [i, j] [i, j+1]
[i+1, j-1] [i+1, j] [i+1, j+1]
current pixel: [0, 0]
[-1, -1] [-1, 0] [-1, 1]
[0, -1] [0, 0] [0, 1]
[1, -1] [1, 0] [1, 1]
*/
- instructions to test whether part 2 works:
- implement peak detection using part 2 using array names consistent with part 1 code.
- beware of potential illusory errors! this step is important and some images might seem correct but are actually incorrect!
- check: output should look like cannypeaks.pgm
- instructions to test whether part 3 works:
- implement hi/lo threshold code and manually enter these thresholds (HI == 100, LO == 35)
- check: output should look like cannyfinal.pgm
- combine code for parts 1, 2, and 3 and generate threshold values using command line percentages
- instructions to test whether part 4 works:
- implement hi/lo threshold "calculator" which automatically calculates hi/lo threshold
- afterwards, you will not have to manually enter hi/lo thresholds, only a percent (use 0.5 or 0.005)
- check: output should look like cannyfinal.pgm (same as part 3)
- psuedocode for part 4:
// part 1: compute gradient vector magnitude
// part 2: compute peaks
// part 4 (comes before part 3!): find hi value based on input percentage automatically
// read percent as input
// computer histogram of scaled magnitudes
for i
for j
(histogram[magnitude(i,j)])++
// compute cutof value (point exceeding percent of all, where hi threshold starts)
cutoff = percent*rows*cols
// move down histogram downwards (right to left) to find hi/low threshold values
for (i = histogram.size; i > 1; i--)
// add total histogram count to area of large magnitudes
areaOfTops += histogram[i]
// when the magnitudes reach a cutoff, break out of loop, found threshold
if (areaOfTops > cutoff)
break;
// assign the threshold values using i from for loop
hi = i
lo = (0.35) * hi
// part 3: double thresholds
- program reads in any picture file (.pgm) and produces three output images;
- magnitude image (magnitude.pgm)
- output for peaks (peaks.pgm)
- final full canny algorithm output, with thresholds (Final Edges.pgm)
- command-line args
- argc: number of args passed
- argv[]: pointer array pointing to each arg passed to program (argv[0] == name of program itself)
$ ./program.out argv[1] argv[2] .... argv[argc] // argv[0] == 'program.out' - compiling C program
# compile with gcc gcc program_name.c -o program_name # execute ./program_name argv[1] argv[2]
- command-line arg for running program:
# canny.c with single threshold
./canny garb34.pgm output_img.pgm threshold
# canny.c with low/high threshold
./canny garb34.pgm output_img.pgm low_threshold high_threshold
# canny.c with three outputs + low/high threshold
./canny garb34.pgm mag_out.pgm low_out.pgm hi_out.pgm low_threshold high_threshold
# output for face05.pgm
./canny face05.pgm cannymag.pgm cannyout1.pgm cannyout2.pgm 40 110
# output for garb34.pgm
./canny garb34.pgm magnitude.pgm peaks.pgm final_edges.pgm 40 110
// output for garb34.pgm
$ ./canny garb34.pgm magnitude.pgm peaks.pgm final_edges.pgm [percent]