Skip to Main Content
In this paper, feedforward neural network trained with Generalized Hebbian Learning rule is proposed to compress grayscale medical images. After training with sufficient sample images, the weights obtained during the training phase will be applied in the compression phase to extract out the principal components that contribute to the most variance in the image. Quantization and subsequently data coding will then be performed on the extracted components before being stored in hard disk. Later at the decompression stage, the original image can be reconstructed by simply multiplying the principal components with the coupling weights. As a comparison, two reference MRI images are utilized, one as the training image while the other as the validation image to verify the generalization capability of the network. Experimental results show that the FFN is able to achieve comparable PSNR against JPEG2000 at 53.36dB and 48.25dB respectively, at a compression rate of 3.0 bits/pixel.