2D FFT (fft2) in SciPy - Time & Space Complexity

Time Complexity: 2D FFT (fft2)

O(n^2 log n)

Understanding Time Complexity

We want to understand how the time to compute a 2D FFT changes as the input size grows.

How does the work increase when we process bigger 2D data arrays?

Scenario Under Consideration

Analyze the time complexity of the following code snippet.

import numpy as np
from scipy.fft import fft2

# Create a 2D array of size n x n
n = 256
data = np.random.rand(n, n)

# Compute the 2D FFT
result = fft2(data)
    

This code computes the 2D Fast Fourier Transform of a square 2D array with size n by n.

Identify Repeating Operations

Identify the loops, recursion, array traversals that repeat.

• Primary operation: The FFT algorithm applies 1D FFTs repeatedly along rows and columns.
• How many times: It processes all n rows and all n columns, each of length n.

How Execution Grows With Input

As the input size n grows, the number of operations grows a bit faster than n squared but much slower than n cubed.

Input Size (n)	Approx. Operations
10	About 1,000
100	About 26,000
1000	About 300,000

Pattern observation: The operations grow roughly like n squared times log n, so bigger inputs take more time but not as fast as cubic growth.

Final Time Complexity

Time Complexity: O(n^2 log n)

This means the time grows a bit faster than the number of data points, but still much faster than simple loops over all points.

Common Mistake

[X] Wrong: "The 2D FFT takes time proportional to n squared because it processes an n by n array."

[OK] Correct: The FFT uses a clever method that is faster than just looping over all points; it adds a log n factor due to the recursive splitting.

Interview Connect

Understanding how FFT scales helps you explain performance in image processing or signal analysis tasks, showing you know how algorithms handle big data efficiently.

Self-Check

"What if we used a non-square input array of size n by m? How would the time complexity change?"

Practice

1. What does the scipy.fft.fft2 function do with a 2D array?

easy

A. It calculates the sum of all elements in the 2D array.

B. It converts the 2D data from the spatial domain to the frequency domain.

C. It sorts the 2D array in ascending order.

D. It normalizes the 2D array values between 0 and 1.

Solution

1. Step 1: Understand the purpose of fft2

   The fft2 function computes the 2-dimensional discrete Fourier Transform, which changes data from spatial to frequency domain.

2. Step 2: Compare options with function behavior

   Only It converts the 2D data from the spatial domain to the frequency domain. correctly describes this transformation. Other options describe unrelated operations.

3. Final Answer:

   It converts the 2D data from the spatial domain to the frequency domain. -> Option B

4. Quick Check:

   2D FFT = spatial to frequency transform [OK]

Hint: Remember fft2 changes space data into frequency patterns [OK]

Common Mistakes:

• Confusing fft2 with sorting or summing functions
• Thinking fft2 normalizes data
• Assuming fft2 works only on 1D data

2. Which of the following is the correct way to import and use fft2 from scipy?

easy

A. import fft2 from scipy
result = fft2(data)

B. import scipy.fft2
result = scipy.fft2(data)

C. from scipy import fft2
result = fft2(data)

D. from scipy.fft import fft2
result = fft2(data)

Solution

1. Step 1: Recall correct import syntax for fft2

   The fft2 function is in the scipy.fft module, so the correct import is from scipy.fft import fft2.

2. Step 2: Check usage correctness

   Calling fft2(data) after import is correct. Other options have wrong import syntax or module names.

3. Final Answer:

   from scipy.fft import fft2 result = fft2(data) -> Option D

4. Quick Check:

   Use 'from scipy.fft import fft2' [OK]

Hint: Import fft2 from scipy.fft module directly [OK]

Common Mistakes:

• Trying to import fft2 directly from scipy
• Using incorrect import syntax like 'import fft2 from scipy'
• Assuming fft2 is a standalone package

3. What is the output of this code snippet?

import numpy as np
from scipy.fft import fft2

arr = np.array([[1, 2], [3, 4]])
result = fft2(arr)
print(np.round(result, 2))

medium

A. [[1.+0.j 2.+0.j]
[3.+0.j 4.+0.j]]

B. [[10.+0.j 2.+0.j]
[ 4.+0.j 0.+0.j]]

C. [[10.+0.j -2.+0.j]
[-4.+0.j 0.+0.j]]

D. [[0.+0.j 0.+0.j]
[0.+0.j 0.+0.j]]

Solution

1. Step 1: Calculate 2D FFT of the array

   The 2D FFT of [[1,2],[3,4]] results in [[10, -2], [-4, 0]] with complex zeros in imaginary parts.

2. Step 2: Round and check output

   Rounding to 2 decimals keeps values same. [[10.+0.j -2.+0.j] [-4.+0.j 0.+0.j]] matches the expected complex array output.

3. Final Answer:

   [[10.+0.j -2.+0.j] [-4.+0.j 0.+0.j]] -> Option C

4. Quick Check:

   fft2([[1,2],[3,4]]) = [[10,-2],[-4,0]] [OK]

Hint: Remember fft2 sums all elements for [0,0] frequency [OK]

Common Mistakes:

• Confusing fft2 output with original array
• Ignoring complex number format
• Mixing up signs in output

4. Identify the error in this code that tries to compute the 2D FFT:

import numpy as np
from scipy.fft import fft2

arr = [[1, 2], [3]]
result = fft2(arr)
print(result)

medium

A. The array shape is invalid for fft2.

B. The input array must be a NumPy array, not a list of lists.

C. The print statement syntax is incorrect.

D. fft2 function is not imported correctly.

Solution

1. Step 1: Check the input array construction

   The arr = [[1, 2], [3]] creates a ragged array (rows have different lengths: 2 and 1), which cannot be converted to a proper NumPy 2D array.

2. Step 2: Identify the error cause

   fft2(arr) fails because np.asarray internally raises ValueError due to invalid shape.

3. Final Answer:

   The array shape is invalid for fft2. -> Option A

4. Quick Check:

   ragged lists cause shape error in fft2 [OK]

Hint: Use consistent lengths in 2D lists before fft2 [OK]

Common Mistakes:

• Creating lists with uneven row lengths
• Not validating array shape before FFT
• Assuming fft2 accepts any list structure

5. You have a 2D image array and want to filter out high-frequency noise using fft2. Which sequence of steps correctly applies this?

hard

A. Apply fft2 to image, zero out high-frequency components, then apply ifft2 to get filtered image.

B. Apply ifft2 to image, zero out low-frequency components, then apply fft2 to get filtered image.

C. Apply fft2 to image, multiply all frequencies by zero, then apply ifft2 to get filtered image.

D. Apply fft2 to image, add random noise to frequency components, then apply ifft2.

Solution

1. Step 1: Understand filtering in frequency domain

   To remove high-frequency noise, transform image with fft2, then zero out high-frequency parts (usually edges of frequency matrix).

2. Step 2: Reconstruct filtered image

   After modifying frequency data, apply inverse fft2 (ifft2) to get back filtered spatial image.

3. Final Answer:

   Apply fft2 to image, zero out high-frequency components, then apply ifft2 to get filtered image. -> Option A

4. Quick Check:

   Filter noise by zeroing high frequencies in fft2 domain [OK]

Hint: Filter noise by zeroing high frequencies, then inverse fft2 [OK]

Common Mistakes:

• Applying ifft2 before fft2
• Zeroing all frequencies removes entire image
• Adding noise instead of removing it