FREQUENCY DOMAIN AND COLOR IMAGE PROCESSING

Source: Digital Image Processing by Rafael C. Gonzalez and Richard E. Woods, PHI, 3rd Edition, 2010
Reference Chapters: Chapter 4 (Sections 4.7 to 4.9), Chapter 6 (Sections 6.1 to 6.3)

1. Basics of Filtering in the Frequency Domain

(Pages 169–172)

In frequency domain filtering, images are first converted to the frequency domain using the 2D Discrete
Fourier Transform (DFT). The transformation separates the image into its frequency components, enabling
manipulation in the frequency space. The filtered result is then converted back to the spatial domain using
the Inverse DFT (IDFT).

This approach is particularly useful when the image processing operation involves detecting or
manipulating periodic patterns or when global filtering is desired.

Advantages:

• Can design filters more easily for global changes.

Steps of Frequency Domain Filtering:

1. Multiply the image by (−1)x+y to center the Fourier spectrum.

M −1 N −1
F (u, v) = ∑ ∑ f (x, y)e−j2π( M + N )
ux vy

x=0 y=0

3. Apply the filter:

G(u, v) = H(u, v) ⋅ F (u, v)

4. Compute the inverse DFT:

M −1 N −1
g(x, y) = Re { ∑ ∑ G(u, v)ej2π( M + N ) }
1 ux vy

MN u=0 v=0

5. Multiply again by (−1)x+y to restore the image origin.

Convolution Theorem:

g(x, y) = f (x, y) ∗ h(x, y) ⟺ G(u, v) = F (u, v) ⋅ H(u, v)

1
This theorem states that convolution in the spatial domain corresponds to multiplication in the frequency
domain, and vice versa.

Applications include satellite imagery enhancement, noise reduction in biomedical scans, and pattern
recognition.

2. Image Smoothing in Frequency Domain

(Pages 172–174)

Smoothing aims to suppress high-frequency components, which are often associated with noise and fine
details, while preserving low-frequency content.

Frequency Domain Distance Function:

D(u, v) = (u − M /2)2 + (v − N /2)2

Ideal Low-Pass Filter (ILPF):

H(u, v) = {
1 if D(u, v) ≤ D0
0 otherwise

- Sharp cutoff, may cause ringing artifacts due to abrupt transitions (Gibbs effect).

Butterworth Low-Pass Filter (BLPF):

1
H(u, v) = 2n
1 + ( D(u,v)
D0 )

- More gradual transition; the parameter n controls the filter's order and roll-off sharpness.

Gaussian Low-Pass Filter (GLPF):

D2 (u,v)
−
2D02
H(u, v) = e

- Smoothest filter; avoids artifacts and ideal for minimizing MSE.

Applications include denoising, background blurring, and smoothing of medical or facial images before
further processing.

3. Image Sharpening in Frequency Domain

(Pages 174–176)

2
Sharpening enhances edges and fine detail by amplifying high-frequency components.

Ideal High-Pass Filter (IHPF):

H(u, v) = {
0 if D(u, v) ≤ D0
1 otherwise

- High-frequency details preserved, but may introduce artifacts.

Butterworth High-Pass Filter (BHPF):

1
H(u, v) = 2n
1 + ( D(u,v)
D0
)

- Gradual transition; better control over noise amplification.

Gaussian High-Pass Filter (GHPF):

D2 (u,v)
−
2D02
H(u, v) = 1 − e

- Smooth frequency emphasis; avoids ringing and is well-suited for edge enhancement.

High-Frequency Emphasis:

He (u, v) = a + b ⋅ HHP (u, v)

G(u, v) = He (u, v) ⋅ F (u, v), g(x, y) = F −1 (G(u, v))

- Where a and b control brightness and sharpness intensity.

Used in CT scans, satellite edge detection, and license plate enhancement.

4. Color Fundamentals

(Pages 394–396)

Color is a result of the human eye’s response to different wavelengths of visible light. The CIE standard
defines color based on three matching functions:

CIE Tristimulus Values:

X = ∫ S(λ)x(λ)dλ, Y = ∫ S(λ)y (λ)dλ, Z = ∫ S(λ)z (λ)dλ

Chromaticity Coordinates:

3
 X Y
x= , y=
X +Y +Z X +Y +Z
- x, y : represent hue and saturation, independent of brightness Y .

Important in scanner calibration, display standardization, and color printing.

5. Color Models

(Pages 397–403)

Color models define how color is represented numerically. They are essential for various applications such
as image acquisition, compression, display, and printing.

RGB to CMY:

C = 1 − R, M = 1 − G, Y =1−B

- CMY is subtractive (used in printers); RGB is additive (used in monitors).

HSV:

C =V ⋅S

X = C ⋅ (1 − ∣(H/60 mod 2) − 1∣)

m=V −C

- Useful for interactive applications like GUI-based color pickers.

YCbCr:

Y = 0.299R + 0.587G + 0.114B

Cb = 128 − 0.168736R − 0.331264G + 0.5B

Cr = 128 + 0.5R − 0.418688G − 0.081312B

- Widely used in JPEG, MPEG video standards.

6. Pseudo-Color Image Processing

(Pages 403–405)

Pseudo-coloring assigns visible colors to grayscale intensities to improve interpretability. This is useful in
thermography, geophysical data visualization, and medical imaging.

4
Intensity Slicing:

Assign different color values to intensity ranges:

I(x, y) ∈ [ak , ak+1 ) ⇒ assign RGB color (Rk , Gk , Bk )

Color Mapping using LUT:

C(x, y) = LUT(I(x, y))

- LUT = Look-Up Table mapping grayscale to color triplets.

False coloring techniques enable clearer interpretation of complex data such as heat maps, elevation
models, or tissue abnormalities.