Visual Inspection Toolbox

NEW PRODUCT
(Left): A mounting board. (Right): A heatmap generated via deep learning highlighting anomalies on the board in red.

Visual Anomaly Detection

Detect, classify, and localize anomalies using deep learning. Train with normal images, identify unknown defects, optimize thresholds, visualize defect regions, and compute metrics.

Localize Screen Manufacturing Defects Using Student-Teacher Anomaly Detector

Documentation | Examples

Soda bottle image with a caliper measurement line between the bottle cap and fill level.

Measurement

Measure distances, angles, and geometric features in calibrated world units. Correct perspective and lens distortion, use caliper tools, and perform precise dimensional inspection.

Automate Calibrated Image Measurement Using Shape Matching

Documentation | Examples

Printed circuit board image with defects localized and classified using a YOLOX object detector.

Object Detection and Counting

Detect, localize, and count objects for visual inspection applications. Build YOLOX-based detectors, generate training data, identify small objects in large images, and count objects using exemplar-based deep learning models.

Train YOLOX Object Detector Using Synthetic Data

Documentation | Examples

Three metal plates labeled NotchedPlate, FlatPlate, and PlateWithHoles, each outlined in blue against a dark background.

Shape Matching

Detect and localize objects in images using edge-based shape matching. Create shape models from template images, then search inspection images for object instances that match those models.

Detect Industrial Parts Using Multi-Class Shape-Based Matching

Documentation | Examples

A synthetic defect inserted into a normal image and labeled as an anomaly.

Synthetic Image Generation

Generate synthetic training data for visual inspection applications. Create labeled datasets by inserting objects into images, enabling efficient training of object detection and instance segmentation models. 

Generate Synthetic Training Images Using Copy-Paste Augmentation

Documentation | Examples

Diagram illustrating the generation of standalone executables, C/C++ code, CUDA code, and ONNX models from visual inspection applications and models.

Deployment and Code Generation

Deploy inspection algorithms and AI models to edge devices using generated C/C++ or CUDA® code. Export models to ONNX for external frameworks. Build standalone inspection dashboards and apps (with MATLAB Compiler). 

Enhancing Quality Control with Machine Learning

Documentation | Examples

Five pills, two labeled as “Good” and three with visible anomalies labeled as “Bad”.

Live Inspection Using Cameras

Use Image Acquisition Toolboxto connect to GigE Vision, GenICam, and other industry-standard cameras, so you can develop and validate visual inspection algorithms using the same images captured by your production system.

Miba AG Creates Deep Learning System for Enhanced Quality Inspection

Documentation | Examples

Product Resources:

Documentation Examples Requirements Release notes
炉の稼働における AI ベースの視覚異常検知

“In the beginning, I was not very familiar with these methods, but I was able to learn a lot about the functional principles from MathWorks software engineers.”

Gernot Reichl, Plansee
View more customer stories

Visual Inspection Toolbox FAQs

Visual inspection is the process of examining manufactured parts, assemblies, or environments to identify defects, verify dimensions, and ensure compliance with quality standards commonly applied in sectors such as manufacturing, automotive, construction, and aerospace.

Automated visual inspection systems use high-resolution cameras and software that support both deep learning and traditional machine vision algorithms such as MATLAB, Visual Inspection Toolbox, and Image Acquisition Toolbox to efficiently detect microscale or nanoscale defects that are difficult for human eyes to pick up.

MATLAB supports student-teacher, PatchCore, FastFlow, and FCDD (fully convolutional data description) through the Visual Inspection Toolbox, each with different characteristics for training speed, model size, and image size handling.

Yes, MATLAB supports unsupervised anomaly detection methods when abnormal images are rare or diverse and supervised object detection methods like YOLOX when there are sufficient abnormal images for training.

The defect detection process includes three main stages: data preparation (including image preprocessing and labeling), the application of classical algorithms (measurement and shape matching) or AI modeling (training anomaly detectors or object detection models), and deployment to embedded hardware or standalone applications.

MATLAB provides functions for noise removal, edge-preserving smoothing, color space conversion, contrast enhancement, and morphology. It also includes the Registration Estimator app to align misaligned images and the Image and Video Labeler app to automate the labeling process using custom algorithms for semantic segmentation or object detection.

Yes, MATLAB enables deployment of deep learning networks and classical algorithms to various embedded hardware platforms including NVIDIA GPUs, Intel and ARM CPUs, and Xilinx and Intel SoCs and FPGAs using code generation frameworks. MATLAB also supports exporting AI models to the ONNX format, enabling integration with external frameworks.

Plansee uses MATLAB to develop an AI-based visual anomaly detection system that automatically identifies damaged furnace carriers before they enter high-temperature furnaces. This helps reduce downtime, prevent furnace damage, and improve production reliability. Musashi Seimitsu Industry uses MATLAB to inspect approximately 1.3 million automotive bevel gear parts per month, and Airbus built an AI model to automatically detect defects in multiple aircraft components before entering service.