Texture Feature Extraction of Image Based on 2D Hilbert-Huang Transform and Multifractal Analysis

Analysis and classification of texture images are significant topics in the field of computer vision. The extraction of texture features from images is mainly applied in areas of object recognition, image segmentation, and image fusion and so on. As images are considered signals, signal processing techniques can commonly be used in image analysis and processing. Traditional signals analysis methods include Fourier transform, short-time Fourier transform, wavelet transform, and others. These methods can effectively analyze stationary signals but cannot effectively analyze image signals. For non-stationary signals, the Hilbert-Huang Transform (HHT) was proposed by Norden E. Huang et al. This method enables accurate analysis of non-linear and non-stationary signals. The HHT can analyze non-stationary and nonlinear signals while interpreting their instantaneous frequency characteristics. This paper extends the theory of HHT to the two-dimensional domain and applies it to image texture analysis. 2D empirical mode decomposition (EMD) algorithm and theory of HHT are introduced first. Then 2D EMD is used to analyze image texture. Thirdly, multifractal spectrum is adopted to describe the image texture, and we give a group experiment on simple classification of natural marble texture. Experimental results show that the proposed method combines the theory of multifractal method and HHT theory to extract features from images, which provide a new way for non-stationary signal fields such as texture image processing.