Oriented Local Binary Pattern (LBPθ): A new scheme for an efficient feature extraction technique

The rich texture information of images places it as one of the powerful means in several fields such as remote sensing, biometrics and image classification, and recognition. The goal of texture analysis is to produce a feature vector (descriptor) of the input image that characterizes the spatial variations within it. Thus, there exist various feature extraction techniques to describe texture information, among them, the Local Binary Pattern (LBP) is widely used to extract the texture features, so, although LBP features are good texture descriptors, they fail, to characterize the image sufficiently. In this work, to enhance the discriminating capability and the classification system accuracy, we propose a new approach for LBP based texture analysis. Thus, the new scheme, called Oriented based LBP (LBP®), is derived from the basic LBP, which use directional neighborhoods and linearly interpolating between the pixel values allows the choice of any direction, θ, to form a bit string, which can constitute a representation of the image. To evaluate the performance of the proposed scheme, an LBP® based palmprint identification system was developed. Experimental results demonstrated that LBP® was effective to improve the biometric identification accuracy and offers better performance and robustness than the basic LBP.