You are on page 1of 2

Barrett CS450 Winter 2011 LECTURES #15-23: Fourier Transforms (Ch.

4)
Key Concepts: 1. Signals can be represented as a sum of sines and/or cosines of varying amplitudes and frequencies (notes) 2. Each point in the frequency domain is a weighted sum of sines and cosines where every point in the signal is a weight. 3. Each point in the signal is a weighted sum of sines and cosines where every point in the Transform is a weight. 4. The Fourier Transform (FT) converts a signal/function to the frequency domain like a histogram of the frequencies contained in the signal 5. The Inverse FT converts the frequency domain representation back to the original signal/function contained in the signal 6. The Discrete Fourier Transform (DFT) converts a discrete digital signal to its frequency domain representation. 7. The DFT can be implemented directly or using matrices (ZfZ) and has ~N2 complexity for 1-D; ~N3 for 2-D 8. The Fast Fourier Transform (FFT) can be implemented directly or recursively and has ~NlogN complexity for 1-D; ~ ~N2logN2 for 2-D 9. The continuous and discrete FTs extend directly to two- and higher dimensions. 10. The Magnitude of the FT tells gives us the combined amplitude of (Real and Imaginary) sines/cosines at a given Frequency; Power Spectrum = Magnitude2 is akin to the amount of energy at a given frequency; The Phase gives us the ratio of Imaginary/Real expressed as an (tan-1) angle. 11. The FT has elegant mathematical properties such as convolution/multiplication, linearity, translation/(linear) phase, gaussian/gaussian, differentiation, etc. 1. Fourier Synthesis and the Frequency Domain: What is Fourier Synthesis? How do we represent/measure spatial frequency? Sketch a high-pass filter. A band-pass filter. A low-pass filter. An input signal can be decomposed into an infinite sum of infinitesimal sinusoids. The FT of a sinusoidal function is and equally space impulse pair. A linear system can be thought of as operation separately on the sinusoidal components of the input signal, which are summed at the output to form the output signal. The Fourier Transform (FT) Show how the FT represents a signal as a weighted sum of sines and cosines. What are the weights? The FT as a "System" is Linear and Shift-Invariant. Explain why. Show it mathematically. The FT of a sum of functions is the sum of their individual transforms (addition theorem). Shifting the origin of a function introduces into its spectrum a phase shift that is linear with frequency and that alters the distribution of energy between the real and imaginary parts of the spectrum without changing the total energy (Shift Theorem). Convolution of two functions corresponds to multiplication of their FTs (Convolution Theorem). Narrowing a function broadens its FT and vice versa (Similarity Theorem). The energy of a function (signal) is the same as that of its FT (spectrum). The Transfer Function of a linear system can be determined as the ratio of its (measured) output spectrum to its (known) input spectrum. If a function of two variables can be separated into a product of two functions of a single variable, then so can its FT. Projecting (collapsing) a 2-D function onto a line at an angle to the x-axis and transforming the resulting 1-D function yields a profile of the 2-D spectrum taken along a line at an angle to the u-axis. Fourier Transform Pairs Explain why the FT of cos(t) is a pair of delta functions at + . Show it mathematically. Use the Sifting Property to show that the FT of (t) is 1. Can you show that the inverse of this is true? Show that the FT of (t) is Sinc2 using the convolution theorem and what you know about the FT of (t). * *** ((t) is simply a triangular function and (t) is a rectangular function .) **** **** *** *** Show that the FT of a gaussian is a gaussian. For a gaussian f(t) with a small , what does F(u) look like? Construct a graphical proof that the FT of a comb function is a comb function. Discrete Fourier Transform (DFT)/Fast Fourier Transform Write the DFT for 1-D. For 2-D. How many points in frequency (Fourier Coefficients) for a discrete signal with n points? Write the DFT using matrix multiplication. For 1-D signals. For 2-D signals. How is the matrix Z obtained? Outline an algorithm for the Fast Fourier Transform (FFT) using recursion. What is the key to the speed of the FFT?

2.

3.

4.

Barrett CS450 Winter 2011


5. Two-Dimensional Continuous and Discrete Fourier Transforms What is spatial frequency? What are the units of its measurement? If we rotate and image, what happens to its spectrum? If we scale and image, what happens to its spectrum? Why does the 2-D FT of a square look like a blurry cross-hair? The point (u,v) in the 2-D frequency domain corresponds to the basis function with frequency u in x and v in y. Under what circumstance can the FT of f(x,y) be written as F(u)F(v)? The FT of s 2-D image is the FT of the rows followed by the FT of the resulting (Fourier Transformed) columns. How does the complexity of the FT of a 2-D image compare with that of a 1-D signal? Fourier Transform Properties Create a convolution filter to smooth a 1-D signal. A 2-D signal. Create one that will smooth even more. Outline a simple algorithm to perform unsharp masking. Explain visually why unsharp masking works. The FT has elegant mathematical properties such as convolution/multiplication, linearity, translation/(linear) phase, gaussian/gaussian, differentiation, etc. The Convolution Theorem states that convolution in the space/time domain is equivalent to multiplication in the frequency domain. Is there a similar relationship for Correlation? What is it?

6.

Examples of Things I should know or Problems I should be able to solve: 1. Know and be able to write/sketch the transform pairs for Rect or (), Sinc(), Triangular or (), (), Step(), Cos(), Sin(), Gaussian or exp{-()2}. 2. Be able to calculate (mathematically using calculus) the forward and inverse transforms of the functions in 1. 3. Be able to compute (by hand or with calculator) the forward or inverse transforms of 1-D or 2-D signals. 4. Derive the Convolution Theorem and explain its significance. 5. Explain the significance of other FT Properties. 6. Answer and explain all questions, properties, theorems, etc. posed in the above outline.

You might also like