Gaussian smoothing uses the 2D distribution as a point-spread function through convolution.The smoothing effect is determined by the standard deviation of the Gaussian. To compute an effective kernel in practice it is common to truncate the filter as about 3 standard deviations from the mean. Here's more details and guidance about Gaussian blur: https://homepages.inf.ed.ac.uk/rbf/HIPR2/gsmooth.htm
keirp
Does this correspond to how blur happens physically? Like in our path tracer?
tanmayghai18
I have the same question as above, I am very curious as to how Gaussian blurring relates to general blurring of renderings and images we have seen throughout the class this semester?
afang-story
It is important to remember to make sure that the values in the filter sums up to 1 to preserve brightness
afang-story
I'm not sure, but I think regarding the questions above there are different types of blurring in those. But an example of something we implemented was simulating depth of field, which is an optical effect. We can create a similar/approximate image that simulates it using a gaussian filter, but its not the "same".
sirejdua
A fun property of the gaussian filter is that its fourier transform is another gaussian filter, which means if we want to blur an image, we can take its fourier transform, multiply that by the original gaussian filter, and take the inverse fourier transform (using the multiplication - convolution theorem).
Gaussian smoothing uses the 2D distribution as a point-spread function through convolution.The smoothing effect is determined by the standard deviation of the Gaussian. To compute an effective kernel in practice it is common to truncate the filter as about 3 standard deviations from the mean. Here's more details and guidance about Gaussian blur: https://homepages.inf.ed.ac.uk/rbf/HIPR2/gsmooth.htm
Does this correspond to how blur happens physically? Like in our path tracer?
I have the same question as above, I am very curious as to how Gaussian blurring relates to general blurring of renderings and images we have seen throughout the class this semester?
It is important to remember to make sure that the values in the filter sums up to 1 to preserve brightness
I'm not sure, but I think regarding the questions above there are different types of blurring in those. But an example of something we implemented was simulating depth of field, which is an optical effect. We can create a similar/approximate image that simulates it using a gaussian filter, but its not the "same".
A fun property of the gaussian filter is that its fourier transform is another gaussian filter, which means if we want to blur an image, we can take its fourier transform, multiply that by the original gaussian filter, and take the inverse fourier transform (using the multiplication - convolution theorem).