Assignments 6.01 and 6.02

Module 6

Assignment

1. Create your own rule

You would have noticed from the density plots, that the rules can be derived just by carefully observing the yellowish region and the axes.

Try to come up with suitable rules by observing the density plots for other color spaces and

compare the performance with the one given in the paper. You might be able to publish your own rule!

The rules in the paper “Module-6.2-Facial-Skin-Detection.pdf” are:

rule1 = r > 95;                  // R>95

           rule2 = g > 40;                  // G > 40

           rule3 = b > 20;                  // B > 20

           rule4 = r > g;                   // R > G

           rule5 = r > b;                   // R > B

           rule6 = abs(r - g) > 15;           // |R-G| > 15

           // max{R, G, B} - min{R, G, B} > 15

           rule7 = (max(max(b, g), r) - min(min(b, g), r)) > 15;

I tried these rules on the images of Obama and Hillary Clinton

From the density plots on RGB color space I deducted the next rules:

rule1 = g > b;                  // g > b

     rule2 = g < b + 60;             // g < b + 60

     rule3 = r > g;                  // r > g

     rule4 = r < g + 100;            // r < g + 100

     rule5 = b > r - 135;            // b > r - 135

     rule6 = b < r - 50;             // b < r - 50

I tried these rules on the images of Obama and Hillary Clinton

From the density plots on YcrCB color space I deducted the next rules:

       rule1 = Cr > 125;                  // Cr > 125

     rule2 = Cr < 175;                  // Cr < 175

     rule3 = Y > 80;                    // Y > 50

     rule4 = Y < 230;                   // Y < 230

     rule5 = Cb > 80;                   // Cb > 80

     rule6 = Cb < 120;                  // Cb < 120

I tried these rules on YcrCB color space with the images of Obama and Hillary Clinton

From the density plots on Lab color space I deducted the next rules:

rule1 = L > 70;                   // L > 70

     rule2 = L < 250;                  // L < 250

     rule3 = a > 125;                  // a > 125

     rule4 = a < 160;                  // a < 160

     rule5 = b > 130;                  // b > 130

     rule6 = b < 165;                        // b < 165

I tried these rules on Lab color space with the images of Obama and Hillary Clinton

From the density plots on HSV color space I deducted the next rules:

rule1 = H > 0;                   // H > 0

     rule2 = H < 22;                  // H < 22

     rule3 = S > 10;                  // S > 10

     rule4 = S < 255;                 // S < 255

     rule5 = V > 40;                  // V > 40

     rule6 = V < 255;                      // V < 255

I tried these rules on HSV color space with the images of Obama and Hillary Clinton

2. Wrinkle Removal

Try to run Blemish removal code on hillary clinton’s image and try to remove the wrinkles by clicking on them. It works to an extent but not very well.

Try to write your own program to remove wrinkles from the face. You might end up making an awesome app!

I modified the program “removeBlemish.cpp” this way:

You push ctrl-key simoultanealy with the left mouse button and move the mouse along one wrinkle from begin to end, all the positions are saved, after that you click the right mouse button and that wrinkle is removed, then choose another wrinkle, when you have finished push the s key to save the image. The routine “Blemish removal” spend 0.3 seconds  in my machine with a radius of 10 in each position, and one wrinkle path use between 150 to 200 positions, that is almost one minute per wrinkle, to diminish the time waiting, one routine of this program remove some positions of the mouse path leaving around 10% of the positions saved, separated a constant distance, then the routine blemish removal, remove the wrinkle in those new positions.

I tried this program with the images of Obama and Hillary Clinton.

BTW, In spite of the image is adapted to the window size and the image showed is deformed, in the program remove wrinkles, the coordinates of the positions of the wrinkles are the coordinates on the image, not on the window.