Noise robust rotation invariant features for texture classification

• Published in: Pattern recognition 2013-08, Vol.46 (8), p.2103-2116
• Main Authors: Maani, Rouzbeh, Kalra, Sanjay, Yang, Yee-Hong
• Format: Article
• Language: English
• Subjects: Applied sciences; Classification; Detection, estimation, filtering, equalization, prediction; Exact sciences and technology; FFT; Fourier transforms; Information, signal and communications theory; Invariants; Local binary patterns; Local frequency descriptors; Noise; Noise robust; Pattern recognition; Rotation invariant; Signal and communications theory; Signal, noise; State of the art; Surface layer; Telecommunications and information theory; Texture; Texture classification
• ISSN: 0031-3203; 1873-5142
• DOI: 10.1016/j.patcog.2013.01.014

This paper presents a novel, simple, yet powerful and robust method for rotation invariant texture classification. Like the Local Binary Patterns (LBP), the proposed method considers at each pixel a neighboring function defined on a circle of radius R. We define local frequency components as the magnitude of the coefficients of the 1D Fourier transform of the neighboring function. By applying different bandpass filters on the 2D Fourier transform of the local frequency components, we define our Local Frequency Descriptors (LFD). The LFD features are added dynamically from low frequencies to high. The features defined in this paper are invariant to rotation. As well, they are robust to noise. The experimental results on the Outex, CUReT, and KTH-TIPS datasets show that the proposed method outperforms state-of-the-art texture analysis methods. The results also show that the proposed method is very robust to noise. ► The paper presents features that are invariant to rotation and are robust to noise. ► The method uses the local frequencies of textures added dynamically from low frequencies to high. ► Local frequency is extracted from a local function located on a circle with radius of R at a pixel (similar to LBP). ► The method outperforms the state-of-the-art methods on the KTH-TIPS, CURet, and Outex datasets. ► The experimental results show the power of the method particularly in extremely noisy conditions.