1. Final Exam Summary - Computer Vision Fundamentals
Most Important Points
- Computer Vision: Image/Video → Information
- Image Processing: Image → Image
- Graphics: Create images
- Three Areas: Measurement, Perception, Search
- Challenge: 3D world projected to 2D images
- OCR: Image text → Digital text
- Applications: Face recognition, medical imaging, autonomous vehicles
- Digital Image: Collection of pixels
2. Final Exam Summary - Digitization & Filtering
Most Important Points
- Digitization: Sampling + Quantization
- Pixel: Smallest image element
- Image Function: f(x,y)
- Storage: B = M × N × k
- Subsampling: Can cause pixelation and checkerboard effects
- Point Processing: Transform each pixel independently
- Histogram: Distribution of pixel intensities
- Histogram Equalization: Automatic contrast enhancement
- Neighbourhood Processing: Uses surrounding pixels
- Filtering: Foundation of image processing techniques
3. Final Exam Summary - Filtering Techniques
Most Important Points
- Filtering: Uses neighbouring pixels.
- Moving Average: Average of neighbours.
- Gaussian Filter: Weighted smoothing using σ.
- Correlation: Apply kernel directly.
- Convolution: Flip kernel first.
- Special Rule: Symmetric kernel → Correlation = Convolution.
- Separability: Gaussian filter is separable.
- Sharpening: Enhance edges and details.
- Unsharp Masking: Original + (Original − Smoothed).
- Median Filter: Best for Salt & Pepper Noise.
- Alpha-Trimmed Mean: Remove extremes then average.
4. Final Exam Summary - Edge Detection
Most Important Points
- Edge: Sudden intensity change.
- Gradient: Measures intensity change.
- Gradient Magnitude: Edge strength.
- Noise: Causes false edges.
- DoG / LoG: Edge detection operators.
- Sigma: Controls smoothing.
- Thresholding: Select edges.
- Canny: Gaussian + NMS + Hysteresis.
- Non-Max Suppression: Produces thin edges.
- Hysteresis: Dual threshold linking.
- Edge Linking: Connects edges.
- Precision & Recall: Evaluation metrics.
5. Final Exam Summary - Binary Images & Morphology
Most Important Points
- Binary Image: Foreground/background only.
- Thresholding: Grayscale → Binary.
- Histogram: Helps choose threshold.
- Dilation: Expands objects.
- Erosion: Shrinks objects.
- Opening: Removes small noise.
- Closing: Fills gaps.
- Structuring Element: Morphology mask.
- Connected Components: Labels objects.
- 4-Connected: No diagonals.
- 8-Connected: Includes diagonals.
- Area: Pixel count.
- Centroid: Object center.
- Bounding Box: Object boundary.
- Circularity: Shape roundness.
6. Final Exam Summary - Colour Image Processing
Most Important Points
- Colour: Light + perception.
- Visible Spectrum: 400–700 nm.
- Rods: Brightness vision.
- Cones: Colour vision.
- RGB: Additive model.
- CMYK: Subtractive model.
- HSV: Hue, Saturation, Value.
- YCbCr: Luminance + chrominance.
- Colour Gamut: Range of colours.
- LAB/LUV: Perceptual colour spaces.
- CBIR: Colour-based retrieval.
- Applications: Skin detection, colourization.
7. Final Exam Summary - Texture Analysis
Most Important Points
- Texture: Repeated patterns.
- Analysis: Segment, classify, synthesize.
- Filter Bank: Extract features.
- LM Filters: 48 filters.
- Feature Vector: Texture description.
- Distance: Similarity measure.
- Clustering: Group textures.
- K-Means: Iterative clustering.
- Textons: Texture primitives.
- Histogram: Frequency distribution.
- Applications: Classification, retrieval.
8. Final Exam Summary - Segmentation Methods
Most Important Points
- Segmentation: Divide image regions.
- Gestalt: Human grouping rules.
- Thresholding: Intensity-based split.
- Clustering: Group similar pixels.
- K-Means: Assign-update-repeat.
- Feature Space: Colour, texture, position.
- Superpixels: Pixel grouping units.
- Graph Cuts: Remove weak links.
- Normalized Cuts: Spectral method.
- Felzenszwalb: Fast graph segmentation.
- SLIC: Superpixel algorithm.
- FCN: Semantic segmentation.
- Mask R-CNN: Instance segmentation.
- SAM: Prompt-based segmentation.
9. Final Exam Summary - Local Features
Most Important Points
- Local Features: Detect and match regions.
- Requirements: Repeatable, distinctive.
- Invariance: Scale, rotation, illumination.
- Corners: Multi-direction intensity change.
- Harris: Corner detector.
- Scale-Space: Multi-scale detection.
- LoG: Blob detection.
- DoG: Fast LoG approximation.
- SIFT: Scale invariant features.
- Orientation: Rotation invariance.
- Descriptor: 128-dim vector.
- Applications: Recognition, tracking, 3D reconstruction.
10. Final Exam Summary - Visual Words & Feature Indexing
Most Important Points
- Feature Indexing: Enables efficient matching in large image databases.
- Descriptor Space: Local features represented as high-dimensional vectors (e.g. SIFT).
- Visual Words: Quantized local descriptors treated as image "words".
- Visual Vocabulary: Collection of visual words created through clustering.
- k-Means: Common algorithm used to form visual vocabularies.
- Quantization: Assign descriptors to the nearest cluster center.
- Textons: Cluster centers representing texture primitives.
- Bag of Visual Words (BoVW): Represents an image using a histogram of visual word occurrences.
- Histogram Representation: Counts frequencies of visual words in an image.
- Cosine Similarity: Measures similarity between BoVW vectors.
- Inverted File Index: Maps visual words to images containing them.
- Sparse Representation: Efficient storage because most words do not occur in every image.
- tf-idf: Emphasizes important words while downweighting common words.
- BoVW Limitation: Ignores spatial relationships between features.
- Improvements: Visual phrases, spatial verification, image sub-grids.
- Image Retrieval: Find images or objects similar to a query image.
- Video Google: Classic object retrieval system using visual words.
- Precision: Relevant Retrieved / Retrieved.
- Recall: Relevant Retrieved / Total Relevant.
- Applications: Image search, object retrieval, large-scale visual databases.