Flood maps are indispensable to regional prioritization and effective resource distribution, and are required by policy makers, insurance firms, and disaster-relief agencies. SAR (Synthetic Aperture Radar) image classification is widely used for flood mapping, although the utilization of image texture has not been well explored. This study proposes a novel SAR-based flood mapping technique that uses optimized Gray Level Co-occurrence Matrix (GLCM)-based texture features, for more accurate flood-extent extraction from COSMO-SkyMed data. The approach involves the extraction of omnidirectional texture features through the use of an optimal window size, followed by independent component transform, which captures most of the information in the first three components and reduces data dimensionality. Flood maps that are derived using a support vector machine classifier were verified against aerial photographs. The presented approach increased the overall classification accuracy by nearly 1.5%.