Fractal measures of image local features: an application to texture recognition

TL;DR

This paper introduces a novel texture classification method that combines fractal measures with local binary patterns, demonstrating competitive results on benchmark and real-world datasets by capturing multiscale information effectively.

Contribution

The paper presents a new approach integrating fractal measures with local binary patterns for improved texture image classification.

Findings

Competitive accuracy on benchmark databases

Effective in real-world plant species identification

Shows the benefit of combining fractal and local features

Abstract

Here we propose a new method for the classification of texture images combining fractal measures (fractal dimension, multifractal spectrum and lacunarity) with local binary patterns. More specifically we compute the box counting dimension of the local binary codes thresholded at different levels to compose the feature vector. The proposal is assessed in the classification of three benchmark databases: KTHTIPS-2b, UMD and UIUC as well as in a real-world problem, namely the identification of Brazilian plant species (database 1200Tex) using scanned images of their leaves. The proposed method demonstrated to be competitive with other state-of-the-art solutions reported in the literature. Such results confirmed the potential of combining a powerful local coding description with the multiscale information captured by the fractal dimension for texture classification.