ACD-Chaos: A Chaos-Blur Image Quality Assessment Dataset Using Henon-Map Chaotic Dynamics

Description

ACD-Chaos is the first image quality assessment (IQA) dataset using Henon-map chaotic dynamics for blur generation. Unlike standard IQA benchmarks (TID2008, TID2013, CSIQ) which use only symmetric Gaussian blur, ACD-Chaos produces asymmetric non-stationary blur patterns that better represent real-world imaging conditions. DATASET CONTENTS: - 34 reference images (ACD1-ACD34) - 170 distorted images (5 blur levels per image) - Human MOS scores from 26 participants - Simulated MOS from Table 2 of associated paper - Subjective study validation files - Blur generation and screening code SUBJECTIVE STUDY: 26 participants rated all 170 images following ITU-R BT.500-13 Single Stimulus protocol on a 0-9 scale. 5 raters excluded after ITU-R BT.500-13 screening (r < 0.75). Final validation: Pearson r = 0.938, Krippendorff alpha = 0.740. ASSOCIATED PAPER: ML-AFFA: A Multi-Level Adaptive Feature Fusion Architecture for No-Reference Image Blur Quality Assessment, with a Chaos-Blur Benchmark and Leakage-Safe Evaluation. Seema Bohra, Mamta Rani, Ravi Raj Choudhary, Sandeep Kumar. Signal Processing: Image Communication, 2026.

Steps to reproduce

STEP 1 - Generate distorted images: Run code/generate_dataset.py. Set reference_images_path to your Reference_Images folder. This produces 170 distorted images in Distorted_Images/ and MOS_Scores/MOS_scores.csv. STEP 2 - Rater screening (optional): Run Code/screen_and_normalize.py Requires 5 Set<N>_combined_long.csv rating files. Produces screening results in Subjective_Study/ folder. Requirements: Python 3.10+ pip install opencv-python numpy scipy pandas

Institutions

• Central University of Rajasthan

  Rajasthan, Ajmer

Categories

Licence