Analysis of scale variations of local features for accurate classification of Emphysema images

The effectiveness of local or global features has recently attracted growing attention in the field of texture image classification and retrieval. The features of the local binary pattern (LBP) for instance, usually lack global spatial information while global descriptors would provide very little local information. This paper proposes two different descriptors to circumvent these shortcomings by providing more information to describe different textural structures of the Emphysema computed tomography (CT) images. The proposed LBP+Multi-fractal Images (LMI) and the rotational invariant LBP+Multi-fractal Images (RLMI) can provide more accurate classification results by using a hybrid concatenation of the local and global features. The experimental approaches are validated for different scales of Emphysema images during the classification process in order to determine the appropriate image size that could yield the maximum classification accuracy. The experimental results show that the descriptors extracted from the combined features considerably improve the performance of the classifiers. The results also indicate that the LMI descriptor outperforms the earlier approaches and demonstrate the discriminating power and robustness of the combined features for accurate classification of Emphysema CT images.