Robust image hashing via visual attention model and ring partition.

Area of Science:

• Computer Vision
• Digital Image Processing
• Information Security

Background:

• Robustness is a critical attribute for image hashing algorithms.
• Existing methods often lack sufficient robustness against significant rotational transformations.
• This limitation hinders reliable image identification and authentication in diverse applications.

Purpose of the Study:

• To develop a novel image hashing algorithm with enhanced robustness against large-angle rotations.
• To improve the discrimination capabilities of image hashing for accurate feature extraction.
• To validate the effectiveness of the proposed hashing method in image copy detection and classification tasks.

Main Methods:

• A visual attention model (Phase spectrum of Fourier Transform - PFT) was employed to generate saliency maps.
• Ring partitioning was applied to the saliency map's LL sub-band for rotation invariance.
• Features were extracted using DWT coefficients on concentric circles, encrypted via a chaotic map, and hashes generated using Euclidean distances.
• Hash similarity was measured using the L1 norm.

Main Results:

• The proposed image hashing algorithm demonstrated significant robustness against digital operations, particularly rotation.
• The method achieved good discrimination capabilities, outperforming several well-known hashing algorithms in classification tasks.
• Simulations on the UCID database confirmed the effectiveness of the hashing technique for image copy detection.

Conclusions:

• The joint exploitation of visual attention and ring partitioning offers a promising approach for rotation-invariant image hashing.
• The developed method provides a robust and discriminative solution for image authentication and copy detection.
• This technique represents a significant advancement in addressing the challenge of rotation robustness in image hashing.