dataset is a massive longitudinal collection of adult face images frequently used for biometric research, specifically in age estimation, gender and race classification, and morphing attack detection. ResearchGate Key Highlights of MORPH-II Massive Scale : It contains approximately 55,134 unique images of 13,000 subjects. Demographic Diversity : The subjects include individuals from African, European, Asian, and Hispanic ethnicities, with ages ranging from 16 to 77 years Longitudinal Aspect

: Because it includes many images of the same individuals arrested multiple times over a five-year span (2003–2007), it is a gold standard for studying how faces age over time in digital systems. "Verified" & Cleaned Versions

While the original dataset is popular, researchers have identified "interesting" inconsistencies—such as self-reported age and gender errors. This has led to the creation of verified subsets University of North Carolina Wilmington | UNCW MORPH-II Inconsistencies and Cleaning : A notable whitepaper from details the process of correcting these errors. MORPH Subgroups and Cleaning : Available on

, this repository provides scripts to clean age metadata specifically to test if face recognition accuracy improves or degrades with age. Train/Val/Test Splits

: Pre-verified splits (typically 80-10-10) are often hosted on platforms like This blog post explores the MORPH II dataset

with labels already provided in CSV format for immediate use in machine learning. Recent "Interesting" Applications Morphing Attack Detection (MAD)

: Researchers use MORPH-II to create "morph" images (merging two people's faces) to see if they can fool biometric systems into verifying both identities. Age Estimation Benchmarking

: It is a primary benchmark for testing AI's ability to predict a person's age within a 5-year margin of error Synthetic Augmentation : New datasets like

use MORPH-II as a "non-synthetic" baseline to compare against high-quality GAN-generated faces. used to clean this data or how to gain access to the official non-commercial version? arXiv:2007.02684v2 [cs.CV] 19 Sep 2020

The MORPH II dataset is one of the most widely used public longitudinal face databases in the world, primarily utilized for research in biometric verification, age estimation, and face morphing attack detection. When researchers refer to a "verified" or "cleaned" version of MORPH II, they are typically discussing refined subsets where metadata inconsistencies—such as self-reported age or race—have been corrected to ensure higher accuracy in experimental results.  Key Features of the MORPH II Dataset 

The standard MORPH II database is a collection of mugshots that provides researchers with critical data for longitudinal studies. 

Scale and Scope: It contains approximately 55,134 unique images from about 13,000 subjects.

Demographic Diversity: The images include male and female subjects from various ethnic backgrounds, including African, European, Asian, and Hispanic.

Age Range: Subject ages vary from 16 to 77 years, allowing for detailed studies on how aging impacts facial recognition over time.

Longitudinal Aspect: The dataset spans from 2003 to 2007, often featuring the same individual across multiple capture sessions.  The Importance of Verification and Cleaning  Verification: TAR at fixed FARs (e

While MORPH II is a benchmark, researchers have identified numerous inconsistencies in its raw data, largely because much of the information was originally self-reported to police departments. 

Data Cleaning: Studies like the MORPH-II Inconsistencies and Cleaning Whitepaper highlight the need to verify age and gender labels to prevent biased or inaccurate research outcomes.

Standardized Protocols: Verified versions often use specific training/testing splits (such as 80-10-10 or 80-20) and automated subsetting schemes to balance racial and gender distributions.

Quality Control: Advanced preprocessing, including face alignment and cropping using tools like DLIB, is standard in verified subsets to ensure uniformity for machine learning models.  Modern Applications in Biometrics 

Verified MORPH II data is essential for developing technologies that can withstand sophisticated biometric threats.  arXiv:2007.02684v2 [cs.CV] 19 Sep 2020

Evaluation metrics

• Verification: TAR at fixed FARs (e.g., TAR@FAR=1e-3, 1e-4), ROC curves, AUC.
• Calibration: EER, DET curves.
• Demographic breakdown: per-group TAR@FAR and false positive rates.
• Age-gap analysis: TAR as function of age difference bins (0–5y, 5–10y, 10+y).

The Future: Beyond MORPH II Verification

As of 2025, while MORPH II remains a historical benchmark, the industry is moving toward larger, privacy-compliant datasets. However, the lesson of verification persists. New datasets like DIVE (Digital IMU Video Environment) and AFAD (Asian Face Age Dataset) now launch with "verified" as a default feature, not an afterthought.

Furthermore, the concept of "verified" is expanding to include:

• Explainability Verification: Can the model explain why it assigned an age?
• Temporal Verification: Are the image timestamps cryptographically signed?
• Synthetic Verification: Is the data generated by a GAN or Diffusion model, and is that provenance verified?

If you meant "FDIA" or "Medical" datasets

There is a possibility of confusion with other datasets:

1. MORPH (Craniofacial Aging): If you are looking for the dataset regarding facial aging (longitudinal study), the paper is "The MORPH Dataset: Longitudinal Face Aging Analysis" by Karl Ricanek and Tamirat Tesafaye.
2. MORPHO: In microbiology, papers titled "Morph II" sometimes refer to bacterial morphology studies, but these are usually specific to isolated experiments and not a famous dataset named "Morph II."

1. Label Noise (Incorrect Age Metadata)

The original collection process involved scraping law enforcement mugshot databases and voluntary photo submissions. Consequently, the metadata—specifically the chronological age and date of capture—is occasionally erroneous. A subject listed as "25" might actually be "27," or the capture date might be misaligned with their birth date. For age estimation models that aim for a Mean Absolute Error (MAE) of under 3 years, a single mislabeled image can skew an entire training batch.

Overview

• Name: MORPH II (often stylized MORPH-II).
• Type: Large longitudinal face image dataset widely used for face recognition, age estimation, and demographic studies.
• Size: ~55,000 images of ~13,000 subjects (male-skewed, strong age-span per subject).
• Common uses: age progression/regression, face verification/identification, bias and fairness research.