Linear Anchored Gaussian Mixture Model for Location and Width Computations of Objects in Thick Line Shape

TL;DR

This paper introduces a novel Gaussian mixture model-based method for accurately detecting the centerline and thickness of thick linear objects in images, outperforming traditional techniques especially in noisy and blurred conditions.

Contribution

It proposes a linear anchored Gaussian mixture model and two EM algorithms, including a modified version with Hessian-based initialization, for improved thick line detection.

Findings

Algo2 outperforms Algo1 in accuracy and speed.

The method effectively handles noisy and blurred images.

The approach provides reliable localization and thickness estimation.

Abstract

Accurate detection of the centerline of a thick linear structure and good estimation of its thickness are challenging topics in many real-world applications such X-ray imaging, remote sensing and lane marking detection in road traffic. Model-based approaches using Hough and Radon transforms are often used but, are not recommended for thick line detection, whereas methods based on image derivatives need further step-by-step processing making their efficiency dependent on each step outcome. In this paper, a novel paradigm to better detect thick linear objects is presented, where the 3D image gray level representation is considered as a finite mixture model of a statistical distribution, called linear anchored Gaussian distribution and parametrized by a scale factor to describe the structure thickness and radius and angle parameters to localize the structure centerline. Expectation-Maximization algorithm (Algo1) using the original image as input data is used to estimate the model parameters. To rid the data of irrelevant information brought by nonuniform and noisy background, a modified EM algorithm (Algo2) is detailed. In Experiments, the proposed algorithms show promising results on real-world images and synthetic images corrupted by blur and noise, where Algo2, using Hessian-based angle initialization, outperforms Algo1 and Algo2 with random angle initialization, in terms of running time and structure location and thickness computation accuracy.