A generalized multiresolution likelihood ratio (GMLR), which can increase the distinction between different signals by fusing their more features, is defined. Multiresolution representation of image characterizes inherent structure of image well, and the GMLR combines each resolution image features with corresponding region features. A spatially variant mixture multiscale autoregressive prediction (SVMMARP) model is proposed to estimate the parameters of GMLR based on maximum likelihood estimation via expectation maximization (EM) algorithm. In the parameter estimation, bootstrap sampling technique is employed. Experimental results demonstrate that the algorithm performs fairly well.