Comparative Analysis of Hough Transform, Fourier Descriptors, And Zernike Moments for Shape Recognition in Noisy Images

Abstract

Problem Statement: In the era of modern digital technologies, image processing is a critical field. Extracting information, detecting objects, and accurately classifying them from noisy or low-quality images hold significant importance. These methods are widely applied in medical diagnostics, industrial quality control, security systems, and remote sensing.

Methodology: This study analyzes three methods based on geometric and invariant features—Hough Transform, Fourier Descriptors, and Zernike Moments—and compares their effectiveness in recognizing shapes in noisy binary images. The experiments were conducted using Python, OpenCV, and Mahotas libraries.

Key Findings: The Hough Transform demonstrated high speed and robustness in detecting traditional geometric shapes.

Fourier Descriptors effectively described shapes based on contours, ensuring invariance to rotation and scaling.

Zernike Moments proved to be the most effective for high-precision recognition but were the most computationally complex method.

General Conclusion: To enhance recognition accuracy, integrating the strengths of each method and combining them with deep learning neural networks represents a promising modern approach.