Texture characterization using local binary pattern and wavelets. Application to bone radiographs

In this paper, we propose a method based on wavelet coefficients associated with 2D and 1D Local Binary Pattern (LBP) descriptors to classify X-ray bone images for bone disease diagnosis. The proposed approach uses two types of algorithms: the “À trous” algorithm that uses B3-spline as a wavelet basis function and the “Mallat” algorithm with the Daubechie wavelet function. The wavelet decomposition is applied to the 2D image and to its projection. Then, the LBP descriptors are performed in both cases. Two approaches were adopted, the first one compares the LBP histograms and the second derives statistical measures from the histograms to form different feature vectors. Experiments were conducted on two populations of osteoporotic patients and control subjects. Results show that the 1D projected field of the 2D images achieves better results for the classification of the two populations.