Digital Image Processing – Image Segmentation (Unit 5), Computer Science / Electronics & Communication Engineering, Academic Year 2024–2025 – Detailed lecture notes and exam-oriented study material

1. Image Segmentation Definition: Partitioning an image into meaningful regions (ROI). Purpose: Simplifies analysis, extracts features, improves recognition. Techniques: Discontinuity-based (edges, points, lines). Similarity-based (thresholding, clustering, region growing). User-interaction-based (manual seed points). Contextual segmentation (spatial relationships). Global-based (histogram, entropy). Learning-based (CNNs, U-Net, Mask R-CNN). 2. Region of Interest (ROI) ROI = selected meaningful regions (e.g., tumor in medical imaging, number plate in surveillance). Purpose: Focus on relevant areas, reduce computation, improve accuracy. Selection methods: Manual (bounding boxes). Automatic (thresholding, edge detection, region growing). 3. Thresholding Converts grayscale images into binary images using intensity thresholds. Types: Global thresholding (single fixed value, e.g., Otsu’s method). Local/adaptive thresholding (varies across regions). Includes histogram-based approaches (unimodal, bimodal, multimodal). Limitations: struggles with noise, uneven lighting, small/dim objects. 4. Region-Based Methods Region Growing: expands from seed pixels. Region Splitting: recursively divides non-homogeneous regions. Region Splitting & Merging: combines both for balanced segmentation. Tools: Quadtree structures for splitting, homogeneity checks for merging. Applications: medical imaging, remote sensing, robotics, surveillance. 5. Line Detection Identifies linear structures (roads, veins, text lines). Types: Horizontal Vertical +45° diagonal -45° diagonal Uses directional masks (matrix filters). 6. Edge Detection Operators Detects sharp intensity changes. First-order operators: Sobel Prewitt Roberts Cross Second-order operator: Laplacian Advanced: Canny edge detection, Kirsch masks (8 directional kernels). Applications: object boundaries, medical imaging, autonomous navigation. 7. Point Detection Identifies isolated points/features. Techniques: Gradient-based, Laplacian of Gaussian (LoG), Harris Corner Detector. Applications: object recognition, motion tracking, image matching.