The mixed raster content (MRC) document-compression standard (ITU T.44) specifies a multilayer representation of a document image. The model is very efficient for representing sharp text and graphics over a background. However, its binary selection layer compromises the representation of scanned data and soft edges. Typical segmentation algorithms that split up the document into layers tend to lift letter colors to the foreground, so that soft edge transitions may not fully belong either to the foreground or background layers, causing “halos” around objects that impair compression performance. We present a method that sharpens the document before compression and softens its edges after MRC-based reconstruction. It builds an edge-sharpening map and estimates the original edge softness at the encoder. The generated map and softness parameters are then used to reconstruct the original soft edges at the decoder. An MRC encoding and decoding scheme based on H.264/AVC and JBIG2 has been used. Experimental results show that, for lower bit rates, the proposed pre-/postprocessing method can improve both subjective and objective compression performance over regular MRC.