(Color online) Fourier spectrum is divided into radial rings and the power spectrum P is averaged over each ring. The frequency f corresponds to the average distance of each ring to the center of the spectrum.
(Color online) Calculus of the fractal dimension of textures. (a) Original texture. (b) Power spectrum. (c) Log-log curve of power spectrum × frequency.
(Color online) Illustration of the richness of texture fractal descriptors. At left, two textures with similar fractal dimensions. At right, the fractal descriptors for each texture and the clear visual distinction between them.
(Color online) A scheme of the proposed method. From up to down, the original texture, the transformed channels, the Fourier curve for each channel, and the final descriptor.
Proposed method generic algorithm.
(Color online) The ability of the proposed descriptors in the discrimination of texture classes. Left, we see texture images from two classes and their respective descriptors. Right, the descriptors are plotted in a same graph, showing visually the high discrimination potential.
(Color online) One image sample from each class of the Vistex dataset. From up to down, left to right: Sand, Tile, Water, Bark, Fabric, Food, and Metal.
(Color online) Some image samples (one from each class) from the USPTex dataset.
Surface visualization of confusion matrices for the classification of USPTex dataset using the compared descriptors. (a) Chromaticity moment. (b) Histogram. (c) Gabor. (d) Fourier.
Percentage correctness rate and respective confidence interval in the classification of Vistex dataset using statistical metrics extracted from MFD curve.
Percentage correctness rate and respective confidence interval in the classification of the tested datasets by the compared descriptors.
Confusion matrices for the classification of Vistex dataset using the compared descriptors. (a) Chromaticity moment. (b) Histogram. (c) Gabor. (d) Fourier.
Article metrics loading...
Full text loading...