ENHANCED FRACTIONAL HARRIS–LAPLACE DETECTOR FOR ROTATION-INVARIANT FEATURE EXTRACTION IN DIGITAL IMAGE PROCESSING

Abstract

Robust feature extraction under geometric transformations remains a fundamental challenge in modern image processing applications, particularly in watermarking, object recognition, and image authentication systems. Conventional feature detectors such as the Harris Corner Detector and the Harris–Laplace Detector (HLD) provide reliable corner localization but degrade in performance under severe rotational transformations due to limited directional adaptability and insufficient scale sensitivity. Advanced approaches such as Scale-Invariant Feature Transform (SIFT) and Affine SIFT (ASIFT) improve invariance properties but often incur higher computational complexity and unstable localization accuracy under extreme geometric distortions. This paper proposes an Enhanced Fractional Harris–Laplace Detector (EFHLD) for robust rotation-invariant feature extraction. The proposed method integrates fractional-order differentiation with the conventional Harris–Laplace framework to improve directional sensitivity, localization precision, and feature stability under varying rotational conditions. Experimental evaluation was conducted using Cameraman, Livingroom, and Peppers images across rotation angles ranging from to . The proposed EFHLD achieved significantly improved repeatability performance compared to conventional HLD, recording mean repeatability values of 43.97%, 70.72%, and 87.15% for Cameraman, Livingroom, and Peppers images, respectively. The proposed method also maintained lower localization errors, averaging approximately 1.33–1.37 pixels, compared to 1.75–1.82 pixels with conventional HLD. The results demonstrate that EFHLD provides stable, accurate, rotation-invariant feature extraction suitable for robust image processing applications.