An adaptive system for compressed video deblocking

• Restoration of heavily compressed video is achieved by an adaptive system.
• Relevant features of the video stream are extracted online.
• A multilayer perceptron is used to control the restoration process.
• The proposed system is shown to outperform previous approaches both quantitatively and qualitatively.

Many state of the art methods for video compression rely on the Discrete Cosine Transform (DCT) and the quantization of its coefficients. Here we present a video deblocking system which is based on the properties of the DCT. Hence it is suitable for MPEG-2 compressed videos, since the MPEG-2 standard belongs to this class of compression formats. The system carries out an online analysis of the video in order to adapt its operation to the characteristics of the incoming video frames. Experiments with benchmark real videos are shown, which demonstrate the performance of our proposal both in quantitative and qualitative terms.