Image Processing Handbook

    Image processing,
    from photons to pipelines

    A working reference for digital image processing: sampling, colour spaces, convolution, geometry, frequency analysis, compression and GPU pipelines.

    Start readingJump to Module 01
    09
    Modules
    28
    Sections
    00
    Runtime dependencies

    Modules

    Each module stands on its own. Read in order for a full pass, or open a single module when a specific problem lands on your desk.

    1. 01Foundations of Digital ImagesFrom photons to arrays: sampling, quantisation, channel layout, colour spaces, and how pixels sit in memory.Analog to DigitalAnatomy of an ImageColour Spaces and ConversionsMemory and Data Structures
    2. 02Point Operations and HistogramsPer-pixel mappings, lookup tables, histogram statistics, equalisation, and threshold selection.Pixel-wise TransformationsHistogram AnalysisThresholding and Binarisation
    3. 03Spatial Filtering and ConvolutionKernels, boundary handling, smoothing families, derivative operators, and sharpening.2D Discrete ConvolutionSmoothing and Noise ReductionEdge Detection and Sharpening
    4. 04Geometry and InterpolationAffine and projective mapping, resampling kernels, and the quality against cost trade-off.Affine and Perspective MappingInterpolation AlgorithmsPractical Applications
    5. 05Frequency Domain ProcessingThe discrete Fourier transform, amplitude and phase, and filters designed in frequency space.Fast Fourier TransformFrequency Filtering
    6. 06Morphology and ContoursStructuring elements, the four base operators, skeletons, contour hierarchies, and watershed segmentation.Binary MorphologyFeature Extraction and Analysis
    7. 07Compression and FormatsEntropy coding, transform coding, modern container formats, and palette quantisation.Lossless CompressionLossy CompressionColour Quantisation
    8. 08High-Performance Web ArchitectureTyped arrays, fragment shaders, worker offloading, and WebAssembly builds of native libraries.Low-Level Pixel ManipulationHardware AccelerationParallelism and WebAssembly
    9. 09Neural MethodsWhere learned models replace hand-written kernels: upscaling, segmentation, and inpainting.Classical Filters against Deep LearningSuper-ResolutionSegmentation and Background RemovalInpaintingDeployment notes