The traditional linear wavelet decomposition has the drawback that it is not optimally suited to represent edge information.And wavelet coefficients are bigger at transition locations than others,which is disadvantage to image compression.To overcome the problems,a new 2-D nonseparable adaptive wavelet transform(AWT) based on lifting scheme has been presented.In this method,the image is adaptively partitioned into smooth and sharp areas according to local information.Thus the method processes edges and homogeneous image regions in different ways through selecting appropriate update and prediction filters.Experiments in JPEG 2000 show that the edges of image are much more reserved,and the dual AWT precedes 5/3 wavelet and 9/7 wavelet.Moreover,the rate-distortion curves exhibit 0.5-2.0 dB improvement in compression performance.