Base on the characteristic analysis of infrared(IR) image target under the strong clutter background, a dim target detecting algorithm was proposed. Background predictor is selected automatically based on Local area of the image texture features, and the estimated image subtracted from the source image gave the residual image, then a improved threshold selection algorithm based on the two-dimensional exponent entropy is used to segment the residual image. The experiment results show that the algorithm is able to quickly extract the target, reduce the false alarm rate, and greatly improve the performance of detecting dim targets.