For gaussian filters, what's the difference between selecting a high standard deviation versus a low one? How would the image change depending on it?
mpotoole
Let's think. A larger standard deviation produces a broader Gaussian function. When convolving this with an image, you will end up averaging pixel values over a larger neighborhood. This will cause your image to look blurrier.
What about the opposite? If we reduce the standard deviation of our Gaussian function, it'll get thinner and thinner, and eventually become an impulse (e.g., [0 1 0]). Convolving your image with this function will end up having no effect (no blur will be applied).
zyl
What is the typical range of values selected for sigma?
mpotoole
It's hard to say---as this depends completely on the application. In the next lecture though, we'll talk about producing Gaussian pyramids, where an image is (essentially) blurred multiple times by Gaussian filters with a range of different sigma values.
For gaussian filters, what's the difference between selecting a high standard deviation versus a low one? How would the image change depending on it?
Let's think. A larger standard deviation produces a broader Gaussian function. When convolving this with an image, you will end up averaging pixel values over a larger neighborhood. This will cause your image to look blurrier.
What about the opposite? If we reduce the standard deviation of our Gaussian function, it'll get thinner and thinner, and eventually become an impulse (e.g., [0 1 0]). Convolving your image with this function will end up having no effect (no blur will be applied).
What is the typical range of values selected for sigma?
It's hard to say---as this depends completely on the application. In the next lecture though, we'll talk about producing Gaussian pyramids, where an image is (essentially) blurred multiple times by Gaussian filters with a range of different sigma values.