How Can I Remove Pixel Noise from ofxKinect Video? - c++

Question

I'm looking for some help figuring out how to remove some low quality pixel noise from a video, that I'm obtaining from an xbox kinect via open frameworks. I'm running logic against "moving" parts of an image, to determine what color is moving the most, and use those regions to also detect the depth of which those pixels are moving. I'm attaching a photo to try to better explain my issue.
http://imago.bryanmoyles.com/xxw80
Of course I know code will be asked for, so I'll post what I have so far, but what I'm looking for more than anything else, is a good algorithm for smoothing out pixelated regions in a photo using C++
for(int y = 0; y < kinect.height; y += grid_size) {
for(int x = 0; x < kinect.width * 3; x += 3 * grid_size) {
unsigned int total_r = 0, total_b = 0, total_g = 0;
for(int r = 0; r < grid_size; r++) {
for(int c = 0; c < grid_size; c++) {
total_r += color_pixels[(y * kinect.width * 3 + r * kinect.width * 3) + (c * 3 + x + 0)];
total_b += color_pixels[(y * kinect.width * 3 + r * kinect.width * 3) + (c * 3 + x + 1)];
total_g += color_pixels[(y * kinect.width * 3 + r * kinect.width * 3) + (c * 3 + x + 2)];
}
}
unsigned char average_r = total_r / (grid_size * grid_size),
average_b = total_b / (grid_size * grid_size),
average_g = total_g / (grid_size * grid_size);
for(int r = 0; r < grid_size; r++) {
for(int c = 0; c < grid_size; c++) {
color_pixels[(y * kinect.width * 3 + r * kinect.width * 3) + (c * 3 + x + 0)] = average_r;
color_pixels[(y * kinect.width * 3 + r * kinect.width * 3) + (c * 3 + x + 1)] = average_b;
color_pixels[(y * kinect.width * 3 + r * kinect.width * 3) + (c * 3 + x + 2)] = average_g;
}
}
}
}
for(int y = 0; y < kinect.height; y++) {
for (int x = 0; x < kinect.width * 3; x += 3) {
int total_difference = abs(color_pixels[y * kinect.width * 3 + x + 0] - rgb[0])
+ abs(color_pixels[y * kinect.width * 3 + x + 1] - rgb[1])
+ abs(color_pixels[y * kinect.width * 3 + x + 2] - rgb[2]);
unsigned char defined_color;
if(total_difference < 40) {
defined_color = (unsigned char) 255;
} else {
defined_color = (unsigned char) 0;
}
color_pixels[y * kinect.width * 3 + x + 0] = defined_color;
color_pixels[y * kinect.width * 3 + x + 1] = defined_color;
color_pixels[y * kinect.width * 3 + x + 2] = defined_color;
}
}
Again, I'd like to reiterate that my code is not the problem, I'm simply posting it here so that you understand I'm not just asking blindly. What I really need, is some direction on how to smooth out pixelated images, so that my averages don't get messed up frame by frame by poor quality.

Answer 1

You can process your image from the camera with some methods from the ofxOpenCV addon. There you will have methods like blur, undistort, erode, etc. Its easy to setup, because its already an addon. Have a look at the openCvExample which should be packed with your openFrameworks. For more information on the mentioned methods, take a look here. If I understand your problem correctly, then a little blur on the image could fix your problem already.