Using visual feature extraction neural network model to improve performance of quadtree based image coding

The authors propose a new technique to improve the performance of quadtree (QT) based image coding through the utilization of a neural network based visual feature extraction model (VFEM). After QT reconstruction is completed, a trained VFEM uses the information contained in the QT reconstructed image to recover the QT reconstruction error. This results in a better quality reconstructed image than the one simply reconstructed from QT representation. Since no extra information other than QT structure itself needs to be transmitted, the VFEM improvement does not increase the coding bit rate. Therefore, a better rate-distortion performance is achieved