A novel embedded image coding method denoted as Cluster Growing Embedding (CGE) is proposed in this paper. Similar to some of state-of-the-art wavelet image coding algorithms, CGE can encode images efficiently by exploiting the self-similarity across subbands in the wavelet domain. By utilizing morphological dilation, CGE forms significance clusters in each wavelet subband, and then exploits the parent-child relationship between subbands to encode the significance map information of one image. A salient new feature of CGE is that it tries to first encode the significance bits in the neighborhood of those already formed clusters. Therefore, CGE is more compatible to the functionality of embedded coding. Meanwhile, CGE is shown to be very suitable to the object-oriented coding framework along with the shape adaptive discrete wavelet transform (SA-DWT).