A large number of remote sensing applications require that images be transmitted across a low bandwidth communication link for human analysis and automatic evaluation. To accommodate the low bandwidth link, compression is often used to reduce the amount of data transmitted. A central part of this compression process is quantization which decreases the entropy in the compressed imagery. While quantization decreases the amount of data it adds nonlinear distortion to the image. This nonlinear quantization noise can severely impair the ability of the analyst and automatic algorithm to identify features of interest in the decompressed data. We develop a wavelet Markov model based approach to detection before quantization to preserve features of interest to the analyst or automatic algorithm.