Notes

Museum exhibit: in the main hole, optics, color, old building. Downstairs thermal imaging camera. microscope. bubble chamber.

Next assiggnment due to 2 weeks. Color and light behaviour and how we perceive that. Something you want to express.

Today we will talk about the idea of areal manipulation. Last week we looked at point manipulation and today we are going to look at neighboorhood of pixels.

The pixels live happily in a neighboorhood, if we look at to the photoshop most of the filters are area manipulation. The important idea is center pixel, it is looking at the neighboorhood pixels and setting the center one. So the manipulation is only change towards the center pixel. Take the average of neighbors and change the center according to that. More blur option looks at more pixels than normal blur. Edge detection is another example of aera manipulation. If you think about the physcology of the pixel a pixel going to the blur it looks and change its value according to its neighboord. The pixel that is going to edge detection going opposite, I wana be as opposite from the neighbooring pixels. So we change the sharpening of the pixels. Unsharp mask does something between blur and edge detection together. IT is more hybird. Most of these effectd are created by method called convolution.

If you go to photoshop and go filters others and see convolution, so center pixel is the pixel in the center on the matrix. We are assigning a pixel and putting a 3×3 pixel :

IF we are trying to blur on a 3×3 matrix. We are changing to every box into one, and then changing the scale to 9 then we get the actual color and only the blur. Add everything up after multiply with the center to these pixels. The farthest pixel is away and it seems unfair that pixel is same amount of influence of the one that is closer. Gausssian blur is better. So we give different importance to the neighbors difference amounts. WE divided by the sum. Take the RGB value of each cell and multipl y by the number there and then divide the sum by the scale.

Sharpening works different. We are trying to shy the center pixel from the rest so what we do is put 9 to the center and -1 to the surrounding ones. The sum of the surrounding ones plus the center ones should give us +1 to see our image.

doing it on the edge of the screen reveals some problem, what we do is we use only the ones that is neighbors and divide with less numbers, wrap it around the image. It is nonsense according to Dan. Two ways you do that are either you wrap it around the image or folding it.

if I know I am doing a 3×3 matrix and I am not manipulating the edge, if you are doing an effect with using every pixel, it resembles a problem. To get an accurate blur in bigger images needs to have more area. This will slow you down, there are few approaches trying to deal with that, some of them other take a smaller image take a smaller matrix. Or using default filters that come with the library.

Using manual edge, highlighting contours that has a good amount of difference then the neighboor. You can do lots of things with aeral manipulation. You can do some stuff based on convolution kernel or something else.

Image here, let;s slice the image from the left to right. Changing the frequencies of the image that are far and that are close.

Taking an image copying into image buffewr and and manipulating that and then copying that to normal buffer and copying that to window.

With the camera we are using camera instead of a file. Only difference this input should be in a loop since we need supply the image of the camera every frame.

For tracking I would go 320×240. For only display 640×480 is ok. Windows machines are faster.