Image Processing Techniques Overview

Study Notes

Image processing involves techniques to enhance image quality, making them more suitable for specific applications.
Reasons for image enhancement include highlighting details, removing noise, and improving visual appeal.
Image enhancement can be viewed as a problem/application-oriented process.
Image enhancement approaches are categorized into spatial domain and frequency domain techniques.
Spatial domain techniques directly manipulate pixels in the image.
Frequency domain techniques process the image based on modifying the Fourier transform of the image.
The spatial domain approach's main categories are intensity transformations and spatial filtering.

Point/Pixel Operations (Intensity Transformation)
- Image Negatives: Inverts pixel values (dark becomes light, and vice versa).
- Log Transformations: Maps pixel values to a logarithmic scale. Useful for images with a wide range of intensities.
- Power Law Transformations: Applies a power function to the pixel values. Useful for adjusting contrast.
- Piecewise-Linear Transformations: Uses piecewise linear functions to map pixel values. Contrast stretching is a common piecewise-linear technique.
- Contrast Stretching: Expands the range of intensity levels in an image to enhance contrast.
Local Operations (Spatial Filtering)
- Spatial Filtering: Performs operations on neighborhoods of pixels. Used for smoothing (noise reduction) and sharpening.
- Examples include smoothing (noise removal) and sharpening (edge detection)
Global Operations
- Histogram Equalization: Adjusts image intensity levels to improve contrast.

Spatial domain enhancement methods are generalized as g(x, y) = T[f(x, y)], where f(x, y) is the input image and g(x, y) is the processed image.
T is an operator on f defined over a neighborhood of point (x, y).
A neighborhood about (x, y) is a sub-image area centered at (x, y). A neighborhood is commonly a square or rectangular sub-image area.

A simplified pixel operation, g(x, y) = T[f(x, y)], is achieved when the operation's output only depends on the input value at the coordinate (x, y).
A neighborhood of size 1 × 1 means the output value g(x, y) depends only on the corresponding input pixel f(x, y).
T represents a gray level/intensity transformation.

Image Negation (or Invert): Inverts the pixel values
Power Law Transform: A pixel-level transformation using a power function.
Log Transform: A pixel-level transformation using a log function.
Piecewise-Linear Transform.: A pixel-level transformation that uses multiple linear segments to map the input to the output grey levels.
Thresholding

A negative image is the total inversion of a positive image.
Light areas appear dark, and dark areas appear light.
Color images' negative inverts colors, such as red to cyan, and blue to magenta.

Contrast stretching is a method to expand the range of intensity levels in an image.
Low-contrast images can result from poor illumination, lack of dynamic range in the imaging sensor, or wrong settings during image acquisition.

Thresholding results in a binary image (black and white) given a mean intensity level (m).

Example Operations:
- Image Smoothing (Noise Removal): Used to blur the image, reducing the impact of noise.
- Image Sharpening (Edge Detection): Used to enhance image edges, creating a sharper image.
Mask Processing or Filtering: A common approach using masks to process or filter images. Masks are small, 2D arrays (e.g., 3x3) whose coefficients in the array determine the process. Smoothing and sharpening operations use different mask coefficients.

Smoothing filters accomplish blurring and noise reduction.
Blurring in preprocessing removes small details before larger object extraction and filling in gaps.
Noise reduction is accomplished via blurring (linear filters) and nonlinear filtering.

The average of the neighborhood pixels under the filter mask replaces each pixel's value.
Called averaging filters or low-pass filters because they suppress high-frequency components (sharp transitions) within the image.
The averaging filters' undesirable side effect is blurring sharp edges

Weights different neighbors differently.
The center pixel is given the highest weight (importance).
The weights decrease as the neighbor's distance from the center pixel increases.