Abstract: An automatic method of simulating low resolution panchromatic image in IKONOS image fusion was described.Firstly,support vector machine(SVM) was used to separate pixels containing low frequency information from those containing high frequency information,which was not suitable to be included in regression coefficients estimating.Secondly,improved Bucket technique was adopted to generate a subset of observations evenly distributed.Finally,low resolution panchromatic image was simulated with parameters achieved by linear regression,and integrated into the Gram-Schmidt spectral sharpening method.Validating experiments were carried out on two datasets of IKONOS panchromatic and multispectral images,visual and quantitative quality judgments show that the proposed method can select evenly distributed pixel observations with low frequency information automatically,which proves its high efficiency,and that the resultant images have less spectral distortion than traditional Gram-Schmidt method does.