Automated Texture Extraction From High Spatial Resolution Satellite Imagery For Land-cover Classification: Concepts And Application

IIi this study, the results of a series of classifications of nine cover types using a texturaVspectral approach are presented. The textu�e extraction method is based on a cooccurrence matnx algorithm. Textura� feature. s were created .fro� a SPOT near-infrared Image usmg four texture mdlces, seven window sizes, and two quantization levels. A supervised classification based on the maximum­ likelihood algorithm has been conducted on the three SPOT spectral bands combined with the four texture images and the three spectral bands combined with each texture image individually. The classification accuracy is measured by the Kappa coefficient calculated from confusion matrices. A factor analysis has been conducted to evaluate the contribution of each variable to the classification accuracy. The addition of a texture image brings a significant improvement. to the classification accuracy of each cover type over the results obtained from the multispectral analysis alone. The window size accounts for 90% of this improvement, while 7% is explained by the band combination, and 3% by the quantization level. There is a window size which optimizes the discrimination of each cover type. The statistics used as texture measures are reduced to a second-level contribution.