Study on classification based on image fusion with curvelet transform

Curvelet transform for image fusion can represent directional edges better than wavelet-based method, and can preserve spectral information more effectively as well. The abilities of capturing texture characteristics of high-resolution images and keeping color information of multispectral (MS) images make the fused images suitable to be classified for higher classification accuracy. To test the effect of curvelet transform on images classification, this paper uses a tunable HIS-Brovey method and curvelet-based method to fuse panchromatic (PAN) and MS IKONOS images respectively at first; then classifies the original MS image and the two fused images, the same training sites of samples are selected during the classification processing; and finally evaluates the classified images. The results show that original MS image, the HIS-Brovey-based fused image and curvelet-based fused image have the overall classification accuracies of 79.80%, 82.83% and 86.87% respectively, among which the curvelet-based classified image obtains the highest accuracy, which indicates that curvelet-based image fusion is more suitable for classification compared with the tunable HIS-Brovey-based fusion method.