Q11Digital Image Processing

Digital image processing is organized as a pipeline in which the output of each stage forms the input to the next, moving from raw pixel data toward meaningful understanding of the image content.

• Image Acquisition: The image is captured using a sensor (CCD/CMOS camera, scanner) and digitized through sampling and quantization to produce an $M\times N$ pixel array, optionally with preprocessing such as scaling.
• Image Enhancement: Manipulates the image to bring out obscured detail or increase contrast, using techniques like histogram equalization and spatial filtering, guided by subjective visual criteria.
• Image Restoration: Improves image quality using objective mathematical models of degradation (blur, noise) rather than subjective judgment, employing filters such as the Wiener filter.
• Color Image Processing: Handles color models (RGB, HSI, CMY) for full-color and pseudo-color processing.
• Wavelets and Multiresolution Processing: Represents images at multiple resolutions for compression and progressive transmission.
• Compression: Reduces the storage or bandwidth required by removing coding, spatial, and psychovisual redundancy (e.g., JPEG).
• Morphological Processing: Extracts image components useful for shape representation, such as boundaries and skeletons, using set-theory operations like dilation and erosion.
• Segmentation: Partitions an image into its constituent regions or objects, one of the most difficult tasks since accuracy determines the success of later recognition.
• Representation and Description: Converts segmented raw pixel data into a form (boundary, region, or feature descriptors) suitable for computer processing.
• Object Recognition: Assigns a label (e.g., "vehicle") to an object based on its descriptors, often using a knowledge base that guides earlier stages as well.

A knowledge base about the problem domain connects and guides all stages, and not every application requires every step; simple tasks may stop at enhancement, while automated inspection systems require the full pipeline through recognition.