Grayscale medical image compression using feedforward neural networks
Woon-Ha Yeo, David F. W. Yap, Tick Hui Oh, D. P. Andito, S. L. Kok, Yih Hwa Ho, M. K. Suaidi · 2011
In this paper, feedforward neural network trained with the backpropagation algorithm is proposed to compress grayscale medical images. In this new method, a three hidden layer feedforward network (FFN) is applied directly as the main compression algorithm to compress an MRI image. After training with sufficient sample images, the compression process will be carried out on the target image. The coupling weights and activation values of each neuron in the hidden layer will be stored after training. Compression is then achieved by using smaller number of hidden neurons as compare to the number of image pixels due to lesser information being stored. Experimental results show that the FFN is able to achieve comparable compression performance to popular existing medical image compression schemes such as JPEG2000 and JPEG-LS.