An unsupervised technique for detecting change area between two SAR images was proposed. The detection process is based on distribution property of the joint intensity histograms and need not distribution hypothesis. The algorithm uses adaptive edge detection to get training data. The joint intensity histograms in different levels are used to decide the membership degree of unlabeled points through Fisher classifier. The fusion model which considers the context relationship and inter-scale information improves the sensitivity. The simulation results of two real SAR images show that the algorithm is effective and has better detection results.