Segmentation of images to bit planes is one of the techniques for scalable image compression. Assuming our source image to be from a uniform quantizer, we categorize its bit planes on the basis of their significance, into two groups called MSB-planes and LSB-planes. The MSB planes contain low-entropy structural information, whereas LSB planes contain high entropy texture information. Due to the different nature of information and entropy of the two groups, they can be coded and reconstructed by different algorithms. The structural nature of MSB planes, make them more compressible at entropy coding stage, whereas the low significance of LSB-planes can be exploited against their high entropy. We realize the later by subsampling of LSB-planes, which in turn requires attention to the close coupling of MSB and LSB planes at the reconstruction stage. We introduce an estimation algorithm for the LSB planes. The reconstructed images are found to be perceptually comparable to the original images. Quantitative comparison also shows significant coding gain in terms of SNR vs. bit rate.