📐HOG

NEW

Also known as: Histogram of Oriented Gradients

A feature descriptor that summarizes local gradient orientations to capture edge structure for object detection.

DESCRIPTION

Histogram of Oriented Gradients (HOG) represents image regions by histograms of gradient directions, emphasizing edge and contour information. It partitions images into cells, computes gradients, and normalizes across blocks to improve invariance to illumination and contrast. HOG is widely used for human and vehicle detection and other object recognition tasks. It is efficient, interpretable, and often paired with linear SVM classifiers. Implementations exist in OpenCV and scikit-image, and it remains a baseline for classical vision tasks despite deep learning’s dominance.

ALTERNATIVE DEFINITIONS

Wikipedia

In computer vision and image processing, a histogram of oriented gradients (HOG) is a feature descriptor used to detect objects.

Concise definition that emphasizes object detection use.

Source

OpenCV

Implementation of HOG (Histogram of Oriented Gradients) descriptor and object detector.

Defines HOG in the context of OpenCV’s implementation and detection capability.

Source

scikit-image

The HOG (Histogram of Oriented Gradients) descriptor counts occurrences of gradient orientation in localized portions of an image.

Highlights localized histogram counting as the core mechanism.

Source

CVPR 2005 (Dalal & Triggs)

The basic idea is that local object appearance and shape can often be characterized rather well by the distribution of local intensity gradients or edge directions.

Original research framing HOG’s intuition and effectiveness.

Source

RELATED TERMS

BRIEF FAST Edge Detection Computer Vision Feature Extraction Pedestrian Detection Support Vector Machine

COMPETING TERMS

SIFT SURF ORB

KNOWLEDGE AREAS

AI Data Science Mathematics

TARGET AUDIENCE

AI Engineer Data Scientist Software Developer MLOps Engineer

EXTERNAL SOURCES

• Wikipedia - Histogram of oriented gradients
• OpenCV - Official HOG descriptor documentation
• scikit-image - HOG tutorial and example
• INRIA / CVPR - Original HOG research paper

TOP QUESTIONS

What is HOG used for in computer vision?

HOG is used to detect objects by capturing edge and gradient structures characteristic of shapes. It remains a strong baseline for tasks like pedestrian detection.

Category: Usage

How does HOG achieve robustness to lighting changes?

HOG normalizes gradient histograms over blocks, reducing sensitivity to global illumination and contrast variations in images.

Category: Technique

How does HOG compare to SIFT or ORB?

HOG is a dense descriptor for detection pipelines, while SIFT and ORB are keypoint-based descriptors. ORB is faster and patent-free; SIFT is more distinctive but heavier.

Category: Comparison

Is HOG still relevant in the deep learning era?

Yes. HOG is lightweight, interpretable, and useful in resource-constrained or classical pipelines, and as a baseline or feature for traditional classifiers.

Category: Relevancy

What are common mistakes when using HOG?

Using inappropriate cell/block sizes and not normalizing properly can degrade performance. Poor pre-processing or ignoring aspect ratios also harms results.

Category: Mistakes

Which parameters matter most in HOG?

Cell size, block size, number of orientation bins, and block stride significantly impact descriptor quality and runtime.

Category: Requirements

Can HOG be combined with machine learning models?

Yes. HOG features are commonly fed to linear SVMs or other classifiers for efficient object detection and recognition.

Category: Application

What datasets are typical for evaluating HOG?

Pedestrian and human detection datasets like INRIA Person and Caltech Pedestrian are commonly used benchmarks.

Category: Example

How does HOG handle rotation?

Standard HOG is not inherently rotation invariant; localized orientation bins help, but significant rotations can reduce performance unless addressed.

Category: Limitations