In　order　to　solve　the　problem　of　low　coding　efficiency　in　the　traditional　multiple　description　image　coding,　this　paper　proposes　a　new　architecture　for　multiple　description　image　coding.　The　proposed　architecture　is　based　on　directional　wavelet　transform　that　removes　redundancy　among　wavelet　domain.　Accordingly,　a　new　interpolation　method　is　used　to　reconstruct　the　image　when　one　of　the　description　is　lost.　The　experiments　show　that　the　proposed　algorithm　performs　better　for　the　compression　rate　and　robustness.　Through　comparisons　with　other　traditional　multiple　descriptions　the　new　architecture　can　obtain　better　quality　of　image　in　unreliable　network　channels.