Effects of image fusion algorithms on classification accuracy

As many sensors are available to acquire higher spatial resolution panchromatic images and lower spatial resolution multispectral images at the same time, image fusion methods have been developed to combine these two types of images multispectral images with higher spatial resolution in order to achieve better interpretation results. Image classification is one type of automated interpretation. As the fusion process alters the pixel value of the original images, it is important to study whether or not the classification accuracy can be improved with the fused images, compared to original images. This paper attempts to analyze the effects of different image fusion algorithms on the classification of fused images and to relate the quality of the fused image to the classification results. Experiments are carried on Quickbird-02 panchromatic and multispectral images over the city of Wuhan (China). From these experiments, it is found that for the unsupervised ISODATA (iterative self-organizing data analysis) classification, classification accuracy has been improved due to the good contrast of fused image; on the other hand, the bad injection of high spatial information into fused image will cause some undesirable classification results.