Coupled pre-/post-processing filters for predictive video coding

摘要

Pre-/post-processing filters obtained from lapped transforms achieve higher coding efficiency by exploiting transform block boundary correlation. Existing approaches however do not explicitly consider the impact of pre-filtering on prediction efficiency, thereby resulting in a lower overall rate-distortion performance. We present a framework where filters are designed with an objective function based on a prediction model. The post-filter is constructed so as to reverse the pre-filter processing at the decoder. We show how the pre-filters thus derived can be refined and local adaptation parameters obtained for better compression efficiency. To further enhance the performance, the mode decision is altered by separating the contribution of luma and chroma components in the rate-distortion model. The overall system performance is measured using a weighted Bjontegaard delta bitrate, showing an overall saving of 2.58% to 6.78% with respect to the “Key Technical Area” (KTA) codec. Additionally, the designed filters significantly reduce the blocking artifacts observed in KTA.