Image Processing Course Summary
Topic 1: Digital Image Fundamentals
1. Definition and Basic Concepts
● A Digital Image is a two-dimensional representation of a picture composed of discrete
elements called pixels. Each pixel has an intensity value (a number representing its
grayscale level or color).
● Analog vs. Digital: Analog images are continuous (as perceived by the human eye),
whereas digital images are discrete (represented by a matrix of numbers).
● Mathematical Representation: An image is represented as a 2D function f(x, y), where
(x, y) are the spatial coordinates and the value of f is the pixel's intensity at that point.
● Levels of Image Processing:
○ Low-level: Primitive operations like noise reduction, sharpening, and contrast
enhancement. Both input and output are images.
○ Mid-level: Involves tasks like segmentation (partitioning an image into regions) and
object classification. The input is an image, and the output can be attributes (e.g.,
object boundaries).
○ High-level: "Understanding" an image, such as object detection and recognition,
often associated with Computer Vision.
2. Image Formation
● Image Model: The intensity f(x, y) is formed by two components:
○ Illumination : The amount of source light incident on the scene.
○ Reflectance : The amount of light reflected by objects in the scene.
○ Formula: f(x, y) = i(x, y) * r(x, y)
● Image Acquisition: The process of converting light energy into a digital signal using sensors
(e.g., single sensors, line sensors, or array sensors as in digital cameras).
3. Mathematical Tools in Digital Images
● Arithmetic Operations: Performed on a pixel-by-pixel basis.
○ Addition: Useful for image averaging to reduce random noise.
○ Subtraction: Useful for detecting changes between two images or for background
subtraction.
○ Multiplication/Division: Used for shading correction or masking (isolating a
Region of Interest / ROI).
● Logical Operations (AND, OR, NOT): Used on binary images for shape manipulation and
masking.
● Geometric Transformations: Modify the spatial arrangement of pixels.
○ Includes scale, rotation, translation, and shear.
○ Requires intensity interpolation (e.g., nearest-neighbor or bilinear) to assign pixel
values to new coordinates.
Topic 2: Image Filtering
1. Definition of Filtering
Filtering is the process of modifying or enhancing an image. There are two main domains:
● Intensity Transformation (Point Processing): The new pixel value depends only on the
original pixel's value. Formula: s = T(r).
● Spatial Filtering (Neighborhood Processing): The new pixel value depends on the original
pixel and its neighbors. It uses a kernel (or mask) that is moved across the image via
convolution or correlation.
2. Intensity Transformations
● Image Negatives: Inverts intensity values (s = L - 1 - r). Useful for enhancing details in
dark regions.
● Log Transformation: s = c * log(1 + r). Maps a narrow range of dark intensities to a
wider range, enhancing details in dark areas.
● Power-Law (Gamma) Transformation: s = c * r^γ. A versatile tool for contrast
manipulation.
○ γ < 1: Brightens the image (enhances dark details).
○ γ > 1: Darkens the image (enhances bright details).
3. Histogram Processing
● A Histogram is a graph showing the frequency distribution of pixel intensity values in an
image.
● Histogram Analysis:
○ Dark Image: Histogram is concentrated on the left side (low intensity).
○ Bright Image: Histogram is concentrated on the right side (high intensity).
○ Low Contrast: Histogram is narrow and concentrated in one area.
○ High Contrast: Histogram is spread out across the full intensity range.
● Image Enhancement Techniques:
○ Contrast Stretching: Maps the existing intensity range to the full available range
(e.g., 0-255).
○ Histogram Equalization: Redistributes pixel intensities to create a flatter, more
uniform histogram, which significantly enhances global contrast.
4. Spatial Filtering
● Smoothing (Low-pass) Filters: Used to reduce noise and blur the image.
○ Box Filter (Averaging Filter): Replaces each pixel's value with the average of its
neighbors.
○ Gaussian Filter: Similar to the averaging filter but uses Gaussian weights, resulting
in a smoother, more natural blur.
● Sharpening (High-pass) Filters: Used to highlight edges and fine details. Based on spatial
derivatives.
○ Laplacian: A second-order derivative operator that is highly sensitive to thin lines
and points.
Topic 3: Morphological Image Processing
1. Basic Concepts
Morphological processing is a collection of techniques for analyzing and manipulating the shape and
structure of objects, typically in binary images.
● Key Components:
○ Object: The foreground pixels being analyzed.
○ Structuring Element (SE): A small binary kernel used as a probe to "test" the
image.
● Fundamental Operations:
○ Erosion: "Shrinks" or "erodes" an object. A pixel remains foreground only if the SE
fully fits inside the object at that pixel's location. Useful for removing small noise and
separating connected objects.
○ Dilation: "Expands" or "thickens" an object. A pixel becomes foreground if the SE at
least hits (overlaps with) the object at that location. Useful for filling small holes and
connecting broken parts.
2. Compound Operations
● Opening: Erosion followed by Dilation. Its effects are:
○ Removes small objects (noise).
○ Smoothes the contours of an object.
○ Breaks thin connections between objects without significantly changing the main
object's size.
● Closing: Dilation followed by Erosion. Its effects are:
○ Fills small holes within an object.
○ Smoothes the contours.
○ Fuses narrow gaps.
3. Advanced Morphological Algorithms
● Boundary Extraction: Obtained by subtracting the eroded image from the original image.
(Boundary = Original - Erosion(Original)).
● Thinning & Skeletonization: Reduces an object to a 1-pixel wide skeletal representation
that preserves the object's topology (connectivity).
Topic 4: Image Segmentation
1. Definition and Goal
Segmentation is the process of partitioning an image into multiple homogeneous and meaningful
regions or segments. Its goal is to simplify the image for easier analysis.
2. Discontinuity-Based Approach (Edge Detection)
Focuses on finding abrupt changes in intensity.
● Gradient Operators: Use first derivatives to find the magnitude and direction of edges (e.g.,
Sobel, Prewitt, Roberts).
● Canny Edge Detector: A multi-stage algorithm considered the gold standard:
1. Gaussian Smoothing: Reduces noise.
2. Gradient Calculation: Finds edge strength and direction using Sobel operators.
3. Non-Maximum Suppression: Thins edges to a single pixel width.
4. Hysteresis Thresholding: Uses two thresholds (high and low) to link strong and
weak edges, creating continuous contours.
3. Similarity-Based Approach
Groups pixels with similar properties.
● Thresholding: The simplest segmentation method.
○ Global Thresholding: Uses a single threshold value for the entire image. Effective
for images with a bimodal histogram.
○ Otsu's Method: Automatically finds the optimal global threshold by maximizing the
inter-class variance.
○ Adaptive Thresholding: The threshold value varies across the image, ideal for
uneven lighting conditions.
● Region-Based Segmentation:
○ Region Growing: Starts with seed points and iteratively adds neighboring pixels that
satisfy a homogeneity criterion.
○ Region Splitting and Merging: Divides the image into blocks and merges adjacent
blocks if they are homogeneous.
● Segmentation by Clustering:
○ K-Means: Groups pixels into K clusters based on features (e.g., intensity, RGB
values). Pixels in the same cluster form a segment.
○ Superpixels: Groups pixels into larger, perceptually meaningful "super-pixels" as a
pre-processing step to simplify segmentation.
● Watershed Segmentation: Treats the image as a topographic map. Segment boundaries
(watershed lines) are the "ridges" separating "catchment basins." Prone to
over-segmentation but powerful when combined with markers.
Dumbdown
1. Introduction to Digital Image Processing
Digital image processing involves using computer algorithms to perform operations on a digital
image to enhance it or to extract useful information.[1][2]
● Analog vs. Digital Image Processing: Analog techniques are used for physical
photographs, while digital methods manipulate digital images using computers.[1]
● Image Representation: A digital image is represented as a 2D matrix of pixels, where each
pixel has a specific value representing its intensity or color.[3]
● Types of Images:
● Binary Image: Consists of only two-pixel values, typically black and white (0 and 1).[1]
● Grayscale Image: A monochrome image with shades of gray, usually with 8 bits per pixel,
allowing for 256 different gray levels.[1]
● Color Image: Typically represented using the RGB (Red, Green, Blue) model, where each
pixel has three values corresponding to the intensity of each color channel.
2. Image Filtering in the Spatial Domain
Image filtering is a technique to modify or enhance an image by changing its pixel values based on
the values of neighboring pixels.[4] This is often achieved through convolution, where a smaller
matrix called a kernel or filter is applied across the image.[5]
● Smoothing (Low-pass) Filters: These are used to reduce noise and blur an image by
averaging the pixel values in a neighborhood.[6]
● Sharpening (High-pass) Filters: These filters enhance edges and fine details in an image
by emphasizing differences in intensity between adjacent pixels.[7]
● Non-linear Spatial Filters:
● Median Filter: This is a non-linear filter that is particularly effective at removing
"salt-and-pepper" noise while preserving edges. It works by replacing each pixel's value with
the median value of its neighbors.[6]
3. Morphological Image Processing
Morphological image processing is a collection of non-linear operations related to the shape or
morphology of features in an image.[8] These operations use a "structuring element" to probe and
transform the image.[9]
● Dilation: This operation adds pixels to the boundaries of objects in an image, effectively
expanding them. It is useful for filling small gaps and connecting disjoint objects.[1][9]
● Erosion: This operation removes pixels from object boundaries, shrinking the objects. It is
useful for removing small, irrelevant objects or noise.[1][9]
● Opening: An erosion followed by a dilation. This operation is effective in removing small
objects and noise while preserving the shape and size of larger objects.[10]
● Closing: A dilation followed by an erosion. This operation is useful for filling small holes in
objects and smoothing their outlines.[10][11]
4. Image Segmentation
Image segmentation is the process of partitioning a digital image into multiple segments (sets of
pixels), often to locate objects and boundaries. The goal is to simplify the representation of an image
into something more meaningful and easier to analyze.[12][13]
● Types of Segmentation:
● Semantic Segmentation: Assigns a class label to each pixel in an image.
● Instance Segmentation: Identifies individual objects within a class.
● Panoptic Segmentation: A combination of semantic and instance segmentation, where
each pixel is assigned a class label and an instance ID.[12]
● Segmentation Techniques:
● Thresholding: This is one of the simplest methods of image segmentation. It creates a
binary image by setting all pixels above a certain threshold value to white and all pixels
below it to black.[12]
● Edge-Based Segmentation: This technique identifies the boundaries of objects by detecting
abrupt changes in intensity. Common edge detection operators include Sobel, Prewitt, and
Canny.[14]
● Region-Based Segmentation: This method groups pixels into regions based on similarities
in properties like color, intensity, or texture.[12][14]
● Clustering-Based Segmentation: Algorithms like K-means clustering are used to partition
the pixels into a predefined number of clusters based on their features.[12]
