NLBP - An Improved Local Binary Pattern for Texture Classifcation using Correlational Features of Neighbouring Pixels

Textural feature extraction concentrates on understanding and capturing the properties of digital images based on the spatial arrangement and relationship between pixels. This technique is applied in numerous applications, such as biometrics, the detection of health conditions, and industrial quality inspections. This paper proposes NLBP (Neighbourhood influenced Local Binary Pattern), an improved Local Binary Pattern based on correlational features between neighbouring pixels. NLBP first divides the grey-level image into overlapping submatrices. The submatrices vary in 3X3, 5X5, and 7X7 windows. Then the centre pixel value of each window is preserved as the threshold value and compared with eight surrounding pixel groups to extract textural information. By substituting the comparison of a single pixel with a set of nearby pixels, the proposed study extracts precise texture information and this algorithm also makes use of various statistical thresholding approaches. The performance of the proposed work was evaluated using KTH and UIUC data-sets under noisy, scaled, illuminated, and rotated conditions which challenges our problem statement. Comparing our work to traditional texture algorithms, it's performance has been improved in various scenarios.