BUILDING DAMAGE DETECTION FROM OPTICAL IMAGES BASED ON HISTOGRAM EQUALIZATION AND TEXTURE ANALYSIS FOLLOWING THE 2016 KUMAMOTO, JAPAN EARTHQUAKE

摘要

In this study, robust building damage detection methodology from optical images observed before and after a disaster is developed based on histogram equalization and texture analysis. The aerial images before and after the main shock of the 2016 Kumamoto, Japan earthquake are used. To avoid the bias of the digital numbers depending on the observation conditions, histogram equalization is applied to the images. The Dissimilarity (DIS) is calculated based on gray level co-occurrence matrix. Normalized DIS index (NDI) from pre- and post-event DIS images is also computed to quantify the change of DIS after the event. Linear discrimination analysis is applied to the post-event DIS and NDI images in order to discriminate collapsed buildings from undamaged buildings using building damage data. The result shows that approximately 75 % of the buildings are correctly discriminated by the proposed model.