percentileNormalizer

R2026b

Create percentile normalizer for anomaly map normalization

Since R2024b

Description

The percentileNormalizer object creates a percentile anomaly map normalizer for a specified anomaly detector using the calculated percentile statistics of normal, or non-anomalous, images in a datastore. Use this object along with the normalize function to normalize anomaly score maps from different detectors to the same range, or to normalize anomaly scores for an anomaly map or a set of anomaly maps to a specified range.

Note

This functionality requires Deep Learning Toolbox™.

Creation

Syntax

normalizer = percentileNormalizer(dsNormal,detector)

normalizer = percentileNormalizer(dsNormal,detector,Name=Value)

Description

normalizer = percentileNormalizer(dsNormal,detector) creates a percentile anomaly map normalizer for a specified anomaly detector, detector, using the calculated percentile statistics of normal, or non-anomalous, images in a datastore, dsNormal.

normalizer = percentileNormalizer(dsNormal,detector,Name=Value) specifies additional options for normalization parameters using name-value arguments. For example, NormalPercentileRange=[0.95 0.99] sets the lower and upper percentile statistics range for normal images to [0.95, 0.99].

Input Arguments

dsNormal — Normal image data

datastore

Normal image data, specified as a datastore that contains image data. Do not include anomalous images in this datastore.

detector — Anomaly detector

studentTeacherAnomalyDetector (since R2026b) object | fcddAnomalyDetector object | fastFlowAnomalyDetector object | patchCoreAnomalyDetector object

Anomaly detector, specified as a studentTeacherAnomalyDetector object, an fcddAnomalyDetector object, a fastFlowAnomalyDetector object, or a patchCoreAnomalyDetector object.

Name-Value Arguments

Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Example: percentileNormalizer(dsNormal,detector,NormalPercentileRange=[0.95 0.99]) sets the lower and upper percentile statistics range for normal images to [0.95, 0.99].

NormalPercentileRange — Normal image percentile range

[0.9 0.99] (default) | vector of form [p₁ p₂]

Normal image percentile range, specified as a vector of the form [p₁ p₂], where p₁ and p₂ are the lower and upper percentile bounds, respectively. The lower and upper percentile values must be in the range [0, 1].

p₁ and p₂ define the slope of the linear mapping from the anomaly score input to the normalized anomaly score values. To learn more, see Linear Mapping for Anomaly Map Normalization.

Verbose — Training progress information display

true or 1 (default) | false or 0

Training progress information display in the command window, specified as a logical 1 (true) or 0 (false).

Properties

PercentileStatistics — Anomaly map percentile values

vector of form [P1 P2]

Anomaly map percentile values, P1 and P2, stored as a two-element numeric vector of the form [P1 P2].

Object Functions

normalize

Normalize anomaly score map using percentiles

More About

Linear Mapping for Anomaly Map Normalization

The percentileNormalizer object maps the upper range of the normal image percentile to the bottom end of the anomaly map output values. The object considers image pixels with anomaly scores higher than the upper percentile of the normal image to be anomalous. p₁ and p₂ are the normal image upper percentile range lower and upper bounds, respectively, and the object bounds the anomaly map output by the lower and upper values v₁ and v₂, respectively. Therefore, this function defines the linear mapping between the anomaly score input x and the normalized anomaly score output y:

$$y=\frac{x(v2-v1)}{(p2-p1)}-\frac{p1(v2-v1)}{(p2-p1)}+v1$$

The upper range values of the normal image percentile, p₂ and p₂, define the slope of the linear mapping from the anomaly score input to the normalized anomaly score values.

The normalized anomaly score from the linear mapping, y, is unbounded. To confine the normalization to a range of values, specify the range using the OutputClipRange name-value argument of the normalize object function.

Version History

Introduced in R2024b

R2026b: Moved to Visual Inspection Toolbox

Starting in R2026b, percentileNormalizer is available in the Visual Inspection Toolbox™ product. In releases R2026a and before, percentileNormalizer is available in the Automated Visual Inspection Library for Computer Vision Toolbox™ add-on.