Overview
Retinex Algorithm is an image processing method that measures local image evidence and returns features, masks, segments, or restored pixels. It is useful when later vision steps need a compact feature, mask, or corrected image.
SpecializedSource checked
Retinex Algorithm
Retinex Algorithm transforms image pixels into features, masks, or restored output.
Quick Facts
Compact metadata for implementation and discovery.
- Category
- Image Processing
- Difficulty
- Intermediate
- Family
- Image Processing
- Tier
- Specialized
- Status
- Verified
- Review
- Source Checked
- Sources
- 2
- Formulas
- 1
- Relationships
- 1
- Code example
- Ready
- Visualization
- Ready
Core Concepts
pixel neighborhoodimage featurelocal responsefilter responseimage mask
Quick facts
Complexity
Runtime and memory behavior at a glance.
Best
O(p)
Average
O(p k)
Worst
O(p k)
Space
O(p)
When to Use
- Use Retinex Algorithm when its image evidence matches the feature, mask, or restoration task.
- Use it when downstream vision code can tolerate its noise and scale assumptions.
When Not to Use
- Avoid when lighting, scale, or noise conditions break the detector's assumptions.
- Avoid when learned or task-specific methods are available and substantially more accurate.
What It Does
- Measures local image structure
- Transforms pixels, regions, or features
- Produces an image, mask, or feature set
Use Cases
- Computer vision preprocessing
- Feature extraction
- Image restoration
Inputs
- Input image
- Kernel, threshold, or model parameters
- Region or feature settings
Outputs
- Filtered image
- Detected features
- Segmentation or transform result
Key Formulas (1)
Retinex response
R = log(image) − log(blur(image))
R = log(image) - log(blur(image))Separates local reflectance from broader illumination.
Pseudocode
- 1.Prepare the image and method parameters
- 2.Evaluate the local response, model, or transform for each relevant pixel or region
- 3.Threshold, rank, refine, or combine the responses
- 4.Return the feature set, mask, filtered image, or restored image
Canonical Example
- Input
- image = small grayscale patch
- Output
- response map, mask, or transformed image
- Explanation
- Retinex Algorithm evaluates local image evidence and returns the pixels, regions, or features selected by the method.
Implementation Notes
- Define border handling and coordinate conventions before tuning thresholds.
- Test on a tiny synthetic image where the expected response is obvious.
Common Pitfalls
- Using fixed thresholds across changing lighting conditions makes output brittle.
- Ignoring border handling changes results near image edges.
Interactive Mechanism Walkthrough
Spatial Primitive & Sweep-Line CanvasShow the input image beside the response map, mask, or transformed output.
Step 1 of 4: Prepare the image and method parameters
image-processingNotes
Retinex Algorithm operates in O(p k) time and O(p) space within the Image Processing family. Key governing relation: R = log(image) - log(blur(image)). Often compared with Gaussian Blur and Thresholding.
Sources (2)
What each reference supports for this entry.
Digital Image Processing
BookImage-processing textbook support for Retinex-style illumination correction and color constancy.
Lightness and Retinex Theory
PaperOriginal Retinex reference for lightness, reflectance estimation, and color constancy.
Open sourceExplore Next
Good next jumps from the relationship graph, ranked for useful discovery.
Relationships
View full relationship pageAlternative toGaussian Blur
Retinex Algorithm extracts or transforms image evidence differently than Gaussian Blur.
Related Algorithms
Tags
Image Processing