Building Damage Level Detection with LBP & PCA

Description

An image analysis project that classifies building damage into minor, moderate, or severe from a single photo. Each image is converted to grayscale, resized, and split into a 5x5 grid, then texture features are extracted with a manually implemented multi-scale Local Binary Pattern (LBP) at radius 1 to 5.

For every grid cell, the normalized LBP histogram is summarized into five statistics (mean, variance, entropy, energy, skewness). The features are standardized and reduced with Principal Component Analysis, where PC1 alone explains 91.82% of the variance.

PC1 values are normalized and mapped to damage categories, and the final label is decided by majority vote across the grid. The clustering is validated with a Silhouette Score of 0.72.