Reduction of blocking effect in transform domain using neural network

We propose a new method using the learning capability of a neural network to remove the blocking effect in block-coded images and show its efficiency. The method adjusts a few frequency coefficients in the transform domain. We use the three layer neural network with the backpropagation algorithm. The neural network learns the correlation between blocks to reduce the blocking effect by adjusting the DCT coefficients in the transform domain. In this proposed method, the neural network has an effect on all coefficients of the dequantized block, though it uses the selected three coefficients (one DC coefficient and two low frequency AC) during the training process. Therefore, it provides a better representation of the human visual property from the viewpoint of blocking effect.