This paper presents a high speed image compression based on the combination of modified self-organizing maps and Back-Propagation neural networks. In the self-organizing model number of the neurons are in a flat topology. These neurons in interaction formed self-organizing neural network. The task this neural network is estimated a distribute function. Finally network disperses cells in the input space until estimated probability density of inputs. Distribute of neurons in input space probability is an information compression. So in the proposed method first by Modified Self-Organizing Feature Maps (MSOFM) we achieved distributed function of the input image by a weight vector then in the next stage these information compressed are applied to back-propagation algorithm until image again compressed. The performance of the proposed method has been evaluated using some standard images. The results demonstrate that the proposed method has High-speed over other existing works.