IEEE Signal Processing Letters · 2013 · 42 citations · 9 references
ComputingComputer ScienceComputer VisionDigital ImagingFeature EngineeringDigital Image ProcessingFeature ExtractionImage ClassificationImage Feature EngineeringImage TextureTexture AnalysisAdditive NoiseRobustnessRotation-invariant Texture AnalysisTexture ClassificationTexture DescriptorsTexture Feature ExtractionTexture FeaturesTexture Recognition
This letter presents a rotation invariant descriptor based on the shearlet transform for texture classification. In the presented method, texture images are first decomposed by the shearlet transform, followed by construction of local energy features. Afterwards, the local energy features are quantized and encoded to be rotation invariant. The energy histograms accumulated over all decomposition levels reflect the different energy distributions and form a new image characteristic. Our method can extract more directional features like orientations in images. Moreover, it is robust with respect to noise. Compared to state-of-the-art texture descriptors, the presented method has comparable classification accuracies on the Outex, Brodatz and CUReT texture databases and shows strong robustness on the databases containing additive noise.
9