A new 2-D fractal dimension estimation based on contourlet transform for texture segmentation

Abstract EN

Texture segmentation has an important role in image processing and pattern recognition applications.

Texture segmentation based on the fractal dimension has better results in analyzing images with high degree of irregularity.

Various methods have been proposed for the fractal dimension estimation and reports have shown that methods based on wavelet transform give more accurate results because of the self-similarity property of fractional Brownian motion.

On the other hand, it was proven that contour let transform can better represent image details compared with wavelet transform.

Therefore, we propose a novel approach of 2-D fractal dimension estimation based on contour let transform.

To do that, after considering the properties of the autocorrelation function of contour let detail coefficients, we estimate fractal dimension using variance of contour let detail coefficients.

Also, to improve the accuracy of segmentation, especially in edges, we utilize an adaptive-size window based on the local high frequency energy.

Moreover, we use a median filter for feature smoothing that leads to better results.

Results obtained by our approach and by other well- known methods demonstrate the superior performance of the new approach.