Similarity measure is usually used to study the similar degree between multisource data

which is the basis of pattern recognition on spatial data.In this paper two kinds of similarity measures are experimentally investigated through some remote sensing image retrievals

they are feature vector based measures and probabilistic measures

accordingly two groups experiments are designed to compare the measures for application to remote sensing image retrieval.From the experiment results we find that in the first group two measures seldom used in the literature perform well

they are χ

2

statistical distance measure and cosine of the angle measure.And in the second group experiments

for computing the similarity degree of two images with their histograms obeying mixture Gaussian distributions

we present a method on the basis of class separability measures according to the K-nearest neighbor rule.The experiment results show that the method has good performance.We believe that the results described in this paper will be of significance in applications to multisource data analysis.