Discrete Cosine Transform with Neural Networks

C.S. Leung (USA)


Neural Networks, DCT


Neural networks offer a powerful solution for image compression apart from the existing technology like JPEG and MPEG. In this research, the input image is subdivided into equal-sized (8×8) blocks and compression is carried out block by block as the tradition Discrete Cosine Transform (DCT) approach. Each pixel is scaled to be an integer between 0 and 255. Each block will transform into a training vector, and the output vector is the DCT of the 8×8 block. A feed-forward neural network is used. The network training function updates weight and bias values according to gradient descent momentum and an adaptive learning rate.

Important Links:

Go Back