An infrared image is decomposed into three levels by discrete stationary wavelet transform (DSWT). Noise is reduced by wiener filter in the high resolution levels in the DSWT domain. Nonlinear gray transformation operation is used to enhance details in the low resolution levels in the DSWT domain. Enhanced infrared image is obtained by inverse DSWT. The enhanced infrared image is divided into many small blocks. The fractal dimensions of all the blocks are computed. Region of interest (ROI) is extracted by combining all the blocks, which have similar fractal dimensions. ROI is segmented by global threshold method. The man-made objects are efficiently separated from the infrared image by the proposed method.