Machine Vision (Introduction): Digital Image Processing
Part of a broader ecosystem of imaging modalities and digital image processing techniques that can be applied anywhere we can form an image be it with visible light, X-rays, gamma rays, microwaves or ultrasound.
Quick summary
Machine vision is what happens when you combine sensing + optics/lighting + digital image processing + decision-making to inspect parts, measure geometry, guide robots, or monitor processes automatically.
Digital image processing (DIP) is the engineering toolbox that turns raw sensor data into images you can trust and then into measurements, features, and pass/fail decisions. Many of DIP’s “classic” techniques were developed in the 1960s, especially for space imagery and early industrial/communications problems, and they still sit underneath modern AI vision systems.
In industry, machine vision means using cameras (and sometimes non-visible sensors), controlled illumination, and algorithms to do tasks like: detect defects, verify assembly, measure dimensions, read codes, and guide robot motion fast, repeatably, and at scale.
For mechanical/mechatronics work, vision is often the missing link between “a mechanism that can move precisely” and “a system that can perceive what actually happened,” closing the loop for automation, quality, and autonomy.
What is digital image processing?
A practical definition
Digital image processing (DIP) is the set of computational methods used to improve, transform, analyze, compress, or measure information from images (where an image is represented digitally as a grid of numbers pixels or voxels). A standard reference text used across engineering programs is Gonzalez & Woods’ Digital Image Processing (4th ed.).
What “digital” really means: sampling + quantization (two knobs engineers actually tune)
A sensor converts energy (light, X-rays, heat, etc.) into an electrical signal. To make it “digital,” you:
- Sample it (turn continuous space into a pixel grid).
- Quantize it (turn continuous intensity into discrete values like 8-bit = 256 gray levels).
Those two steps create your basic engineering tradeoffs:
- Higher resolution vs. more data/compute
- Higher bit-depth vs. more bandwidth/storage
- Faster frame rate vs. lower exposure/SNR
Where DIP sits inside a machine vision pipeline
A common industrial pipeline looks like:
- Image acquisition (sensor + lens + lighting + triggering)
- Pre-processing (denoise, deblur, normalize lighting, correct distortion)
- Segmentation (separate object from background)
- Feature extraction (edges, corners, blobs, templates, 3D points, etc.)
- Measurement / classification / decision (metrology, defect detection, ID)
- Action (reject part, guide robot, log trend, stop line)
Even if you later use deep learning, steps (1)–(3) are still where many real systems succeed or fail.
The origin of digital image processing
Why DIP “took off” in the 1960s
Many foundational DIP techniques were developed in the 1960s at organizations such as Bell Labs, JPL, MIT, and others driven by needs like satellite/space imagery, standards conversion for transmitted photos, medical imaging, character recognition, and enhancement of noisy images.
A concrete early story: NASA/JPL and lunar images
NASA’s Jet Propulsion Laboratory (JPL) documents early work where engineers used computers to enhance and reconstitute space images. JPL describes computer enhancement improving lunar/Martian imagery resolution and notes this work ramped up around the time of Ranger VII (July 1964) lunar imaging and related efforts.
The big idea was (and still is): if you understand the sensor and noise/distortion sources, you can use computation to recover a much more useful image than the raw data suggests.
Examples of fields that use digital image processing
1) Manufacturing automation (classic “machine vision”)
On factory lines, machine vision is used for:
- Automated inspection (scratches, missing features, wrong part)
- Dimensional measurement (gauging edges, holes, spacing)
- Robot guidance (pick-and-place, bin picking, alignment)
- Verification (orientation, seating, presence/absence)
Industrial vendors commonly highlight micron-scale measurement capability when the optics/lighting/calibration are engineered correctly.
2) Medical imaging
CT, nuclear medicine (PET/SPECT), MRI, and ultrasound all rely heavily on reconstruction and image processing to turn sensor measurements into clinically useful images.
3) Remote sensing + mapping (Earth, oceans, weather, agriculture)
Remote sensing uses image processing for calibration, atmospheric correction, multi-band fusion, classification (e.g., vegetation/land cover), and change detection. NASA notes that many passive remote sensing systems operate in visible/IR/thermal IR/microwave, and active sensors (radar) are often in microwave bands.
4) Astronomy and high-energy science
Radio, infrared, X-ray, and gamma-ray telescopes produce data that must be processed into images (often through heavy reconstruction), because the “camera” may be an array, a scanning instrument, or a photon-counting detector rather than a normal lens + CMOS sensor.
Gamma-ray imaging
What it is (and why it’s different)
Gamma rays are the highest-energy region of the electromagnetic spectrum. NASA’s spectrum overview places gamma rays above X-rays in energy.
Because gamma rays are extremely penetrating and ionizing, gamma-ray imaging is less about “taking a picture with a lens” and more about detecting individual photon interactions in specialized detectors (often with scintillators and photodetectors) and reconstructing where they came from.
Medical example: PET and SPECT (nuclear medicine)
NIBIB explains that in PET, positrons annihilate with electrons, producing two coincident gamma-ray photons emitted in opposite directions; detectors measure these events to form images.
RadiologyInfo also describes PET as nuclear medicine imaging where the tracer gives off gamma rays that are detected by a special camera system.
SPECT is another major gamma-based modality; nuclear medicine systems such as gamma cameras/SPECT/PET detect emitted energy and computers create the images.
Industrial example: radiographic inspection using gamma sources
In industrial nondestructive testing (NDT), radiographic inspection uses X-rays or gamma rays to “see inside” materials (especially welds and critical structures).
Here, image processing supports defect visibility (contrast enhancement), measurement, archiving, and increasingly automated flaw detection.
X-ray imaging
The core physical principle
X-ray imaging leverages differential attenuation: dense/high-Z materials absorb more X-rays than low-density materials, producing strong contrast.
Medical CT (computed tomography)
NIBIB defines CT as a computerized X-ray imaging procedure where a narrow beam of X-rays rotates around the body and a computer reconstructs cross-sectional “slices.”
The CDC similarly describes CT as combining a series of X-ray images into a 3D picture (with higher radiation exposure than a typical single X-ray).
Industrial X-ray CT (metrology + defect detection)
Industrial CT is widely used to inspect internal geometry/porosity without cutting parts—particularly relevant to castings and additive manufacturing. NIST has published work specifically on using X-ray computed tomography for inspecting internal features in additively manufactured metal parts.
This is a strong mechanical engineering example of imaging as a “virtual sectioning tool,” and DIP is essential for reconstruction, artifact reduction, segmentation, and dimensional extraction.
Imaging in the UV band / visible and IR bands / microwave / radio band
Below is a “mechanical engineer’s interpretation” of what each band tends to reveal and why it changes the image processing problem.
First: where these bands sit in the electromagnetic spectrum
NASA summarizes the electromagnetic spectrum ordering (high to low energy) as:
Ultraviolet (UV): surface effects + fluorescence tricks
UV is widely used in inspection because it can make certain materials fluoresce, turning tiny cracks/contamination into high-contrast features.
- NDT example (very relevant to mechanical engineering): UV light is used in fluorescent penetrant inspection workflows to reveal surface-breaking flaws. A quality/inspection trade source notes UV inspection methods are staples in NDT, including fluorescent penetrant processes.
- UV exposure is also tightly tied to atmospheric ozone behavior; NASA explains ozone’s role in absorbing UV and discusses UV impacts on life important context for why UV imaging is often controlled/limited and why some UV observations are space-based.
Image processing angle: UV inspection often benefits from background suppression, thresholding, blob/region analysis, and careful control of ambient light (so contrast is dominated by fluorescence rather than reflections).
Visible (VIS): “default” machine vision
Visible light is the workhorse for most industrial machine vision because:
- sensors are cheap and high quality,
- optics and lighting are mature,
- humans can directly interpret results.
Industrial machine vision references emphasize that success depends heavily on lighting, lenses, and controlled imaging conditions because “what the camera can’t see, you can’t inspect.”
Image processing angle: classic DIP lives here filtering, edge detection, geometric calibration, morphology, template matching, OCR, and (in modern systems) deep learning.
Infrared (IR): temperature, thermal patterns, and “seeing” heat
NASA notes that humans can’t see infrared but can detect it as heat, and highlights “thermal infrared” wavelengths (often cited around 8–15 µm) as useful for studying thermal energy radiating from Earth.
Mechanical/mechatronics examples:
- Predictive maintenance (bearings, motors, electrical panels) through hot-spot detection and trend monitoring.
- Thermal process monitoring (curing, welding heat-affected zones depending on setup).
Image processing angle: IR data often needs emissivity/reflectance awareness, normalization, and trend-based analytics (changes over time can matter more than absolute temperature in a single frame).
Microwaves: radar imaging and seeing through clouds (often)
Microwaves are a key band for radar-based imaging. NASA explains that radar is an active remote sensing system: it sends a microwave pulse and measures the reflected energy.
NASA Earthdata describes Synthetic Aperture Radar (SAR) as sending out pulses of energy and recording what comes back, producing images based on interactions with the surface.
Engineering examples:
- Earth observation (flood mapping, deformation via InSAR, ice monitoring)
- Through-weather/through-cloud imaging (microwave advantages in many conditions)
Image processing angle: SAR and radar imaging are heavily computational: calibration, speckle/noise handling, phase processing (InSAR), georectification, and classification.
Radio: long wavelengths, big antennas, and interferometry
NASA’s radio waves page notes that because radio waves are much longer than optical waves, radio telescopes are built differently than visible-light telescopes (often large dishes/arrays).
NASA’s Madrid Deep Space Communications Complex (MDSCC) explains that radio astronomy uses large antennas and, crucially, radio interferometry/aperture synthesis, where multiple linked telescopes improve angular resolution.
Image processing angle: radio “images” are frequently reconstructions from arrays (Fourier-domain processing, deconvolution, calibration), which is conceptually closer to CT reconstruction than to a smartphone camera.
Examples in which other imaging modalities are used (not just EM-wave bands)
A useful mental model: “imaging” means mapping a physical property over space, not necessarily using electromagnetic radiation.
Ultrasound (mechanical waves)
NIBIB explains that ultrasound imaging uses echoes: by using the speed of sound and echo return times, the system computes distances to tissue boundaries to form images.
Mechanical analogy: this is like active sonar time-of-flight + reflection strength → reconstruction.
MRI (magnetic resonance; RF signals in strong magnetic fields)
The FDA describes MRI scanners as using strong magnetic fields and radiofrequency (RF) energy to make images.
MRI is a great reminder that “imaging” can be physics-heavy and reconstruction-heavy even when no ionizing radiation is used.
LiDAR / laser scanning (3D geometry)
NASA Earthdata explains lidar as using laser pulses to measure distance, enabling accurate 3D measurements (point clouds).
For mechatronics: LiDAR is huge in robotics (mapping, navigation) and industrial metrology (3D inspection), and processing includes point cloud filtering, registration, meshing, and dimensional extraction.
Industrial NDT modalities (common in mechanical engineering)
Beyond radiography (X-ray/gamma), welds and structures are inspected using other NDT methods such as ultrasonic testing; NDT references often group these as complementary tools depending on defect type and access.
Summary
- Digital image processing is the engineering toolkit that turns sensor outputs into usable images and measurements.
- DIP’s modern roots trace strongly to the 1960s, including major work at places like JPL (space imagery enhancement) and other research labs.
- “Images” are not limited to visible light. Real engineering systems use modalities across the EM spectrum:
- Gamma (PET/SPECT; industrial radiography)
- X-ray/CT (medical + industrial internal inspection)
- UV (fluorescent inspection)
- IR (thermal patterns, maintenance)
- Microwave (radar/SAR)
- Radio (astronomy, interferometry)
- Many important imaging systems are non-EM (ultrasound, MRI physics, LiDAR geometry), and they still depend on DIP for reconstruction, enhancement, segmentation, and measurement.
(Value-add) A simple rule-of-thumb for choosing an imaging modality in engineering
If you remember one thing: choose the band/modality that makes your target defect/feature produce the biggest physics-based contrast, then use image processing to make that contrast measurable and robust.
- Want internal voids/porosity? → X-ray CT
- Want surface cracks that fluoresce? → UV inspection
- Want abnormal heating in equipment? → IR thermography
- Want all-weather mapping? → microwave radar/SAR
That’s the heart of machine vision engineering: physics first, algorithms second, and deployment constraints (cost/safety/speed) always.
- Machine vision - Wikipedia
- GeeksforGeeks
- Jpl
- Machine Vision Basics
- Mathworks
- Digital image processing - Wikipedia
- Computed Tomography (CT)
- Earth Observation Data Basics | NASA Earthdata
- Radio Waves - NASA Science
- The Electromagnetic Spectrum (EMS) - NASA Science
- Nuclear Medicine
- PET/CT
- General Nuclear Medicine
- Radiographic Inspection (RT) | CWB Group
- etc.
