My CV note.

>> im = imread('perpers.png');
>> imgreen = im(:,:,2); %matlab index start at 1. so green channel is 2.
>> imshow(imgreen)
>> line([1 512],[256 256],'color','r');
>> plot(imgreen(256,:));
>> disp(size(im));
>> disp(class(im)); %uint8
>> 

Multiply by a scal function 函数的写法

function result = scal(img, value)
	result = value .* img;
endfunction

产生Gaussian noise

noise = randn([1, 100]);
[n, x] = hist(noise, [-3, -2, -1, 0, 1, 2, 3]);

n是数量，x是bin。

disp([x; n]);

 	-3    -2    -1    0     1      2     3
    1     4    27    35    26     6     1

noise = randn([1, 10000])
[n, x] = hist(noise, linspace(-3, 3, 21));
plot(x, n);

randn-> Standard Normal Distribution

rand -> uniform distribution

randi-> random integers

##3 Effect of sigma on Gaussian Noise noise = randn(size(im))*sigma;


adding a noise function to an image function

noise = randn(size(im)).*sigma;
output = im + noise;

output was just the image + noise worry to the range of the image: image [0 255] sigma=5 should plausible, if image [0 1] that is not plausible.



replace each pixel with an average of the values of the pixels neighborhood, this is referred to as a moving average. 最终得到smooth后的结果。

1 The "true" value of pixels are similar to the true value of pixels nearby.

2 The noise added to each pixel is done independently.

non-uniform weight 与数值相邻近。

To do the moving average conputation the num of weights should be Symmetric and Odd-makes it easier to have a middle pixel.

image - 2D

save the averaging window size is .

-->Uniform weight for each pixel.

-->loop over all pixels in neighborhood around image pixel F[i,j].

Correlation filtering - non-uniform weights

Now generalize to allow different weights depending on neighboring pixel's relative position: This is called cross-correlation, denoted .

H[u,v]--> None-uniform weights The filter "kernel" or "mask" H[u,v] is the matrix of weights in the linear combination.

what went wrong with the output?

squares are not smooth, filters are blurring of filtering an image to make it smoother doesn't seem right.

To blur a signle pixel into a "blurry" spot, we would need to need to filter the spot with a something that look like a blurry spot - higher values in the middle, falling off to the edges.

non-Gaussian filter you can see those shape edges.

Gaussian filters are referred to as exponentials. The complete formular is:

The Gaussian we just talked about is what's referred to as isotropic(各向同性) and it basically had 1 parameter $\sigma$, the bigger the more blur. the filter: the size of matrix $(3\times 3)$,$(5\times 5)$,$(11\times 11)$.

The bigger kernel, it's smoother, it's going to work better.->actually mean a bigger $\sigma$. That's what you care about is the size of the sigma, not actual kernel.

MATLAB 用surf函数画出来的图 叫surf 三维着色表面图、三维表面图、表面图。imagesc(A)将矩阵A中的元素数值按大小转化为不同颜色，并在坐标轴对应位置处以这种颜色染色。 surf() --> surface; imagesc() --> show image