Image Processing and Edge Detection

• The algorithms below work with black and white images (gray scale)
  and highlight the edges in the image.
  This is a first step in identification of objects in an image
• These examples are greyscale, but the ideas work in colored images too.
• The techniques here are horizontal and vertical differencing, Roberts, and Sobel.
• You need to read in the "pgm" file, read in the number columns and rows of the figure.
• Next, process through the matrix of pixel values for the image,
  and change each pixel value based on the surrounding pixels
• A pixel is turned brighter based on the rate of change of it and its neighbors.
  A higher rate of change (this happens on an edge) is colored brighter
  than a low rate of change (not an edge)

  1. Copy the image file:
    - Use Gimp or a browser to view fig1.pgm
    - This is a "pgm" format image file
    - Here's the data file for fig1.pgm, this is a P2 format file with 97 columns, 128 rows, and max pixel value 255 (white)
     
  2. Perform the following edge detection routines:
     A. Horizontal differencing: 
          image2[row][col] = abs(image[row][col+1] - Image[row,col])
  	Note: if (col == cols-1) image2[row][col] = 0;
     	    Set the pixels in the last column equal to 0
  	
     Your result should look like: 
     
  
     
  
     B. Vertical differencing:
          image2[row][col] = abs(image1[row+1][col] - image1[row][col])
  	Note: if (row == rows-1) image2[row][col] = 0;
                Set the pixels in the last row equal to 0
  
         Your result should look like: 
         
  
         
  
     C. One version of Robert's cross:
  
  	image2[row][col] = abs(image1[row][col-1] - image1[row][col+1]) +
                        abs(image1[i-1][j] - image1[i+1][j]);
  	Note: if (row == 0 || row == rows-1 || col == 0 || col == cols-1)                    image2[row][col] = 0;
  	   Set the first and last rows and cols to 0
  
  
        Your result should look like:
        
  
        
  
      Robert's Cross version using sqrt and sqr:
   	image2[row][col]= floor(sqrt(sqr(image1[row][col-1] - image1[row][col+1])  + 
  	sqr(image1[row-1][col] - image1[row+1][col])));
  
  	Note: if (row == 0 || row == rows-1 || col == 0 || col == cols-1)                    image2[row][col] = 0;
  	   Set the first and last rows and cols to 0
        
  
        
  
  
     D. Sobel's operator:
        For this pattern around pixel e:
           a  b  c
           d  e  f
           g  h  i
  
          dx = c + 2f + i - a - 2d - g
          dy = a + 2b + c - g - 2h - i
          
          image2[row][col] = (int) sqrt (sqr(dx) + sqr(dy))
          Use (int) or floor(x) to truncate the sqrt.
          If the value is > 255, reset it to 255 so that 255 is the max value
  	
          Set the first and last rows and cols to 0
  
  	Your result should look like: 
  	
  
  	
  
  
  3. Save each of the previous edge detected images under different file names.
  
         Fig1HDiff.pgm - Horizontal Differencing
         Fig1VDiff.pgm - Vertical Differencing
         Fig1Roberts.pgm - application of the Robert's cross pattern
         Fig1Sobel.pgm - application of the Sobel operator pattern
  
     Compare the strength of each edge detection algorithm.
     Is a thicker line like Sobel's result necessarily better than a thinner, sharper line?
  
  4.  For Supercompting Applications: Try one or more of these edge detection methods using parallel processes.