Abstarct: The increasing growth of digital data in recent years has led to increased attention to compression. The emergence of the phenomenon of compression of image with the neural network introduces us a promising way of developing new effective compression techniques. The main purpose of this phenomenon is to reduce data redundancy. So far, good improvements have been made to image and video compression techniques, and improvements have been made to conserve memory for the transmission of visual information. In this thesis, the Deep Neural Network (DNN) and Noise Reduction Filter for compression of natural color images are proposed. Simulation results show that our proposed method is effective in reducing image size and compression rate, as well as improving the image quality compared other methods. For Instance, in the prposed method which uses noise-cancelling filter, the PSNR value has increased from 22.66db to 24.3db for lena image and from 19.73db to 22.10db for mandrill image compared to image compression method baxsed on neural network(baxsed on similar pixels) at an equal compression ratio. The prposed method has enhanced both image quality and the values.