An image coding scheme based on multiple description scalar quantization (MDSQ) is presented in this paper. The proposed approach targets to improve coding efficiency for networks which have tendency to pocket losses. In this work, it is aimed to estimate quantization indexes of lost descriptions at the decoder in least squares sense using optimal filter coefficients obtained at the encoder. Thus, tire filter coefficients obtained at the encoder are also need to be transmitted in addition to descriptions. Experimental results show that the proposed approach provides better performance compared to conventional multiple description scalar quantization approach.