Method and associated device for filtering digital video images

What is this patent about?

’891 is related to the field of digital video processing, specifically the reduction of visual artifacts in block-based video coding. In standard compression schemes like H.263, images are divided into discrete blocks for transformation and quantization, a process that frequently introduces discontinuities at the edges of these blocks. These discontinuities, known as blocking artifacts, degrade image quality and can propagate through subsequent frames when used in motion-compensated prediction.

The underlying idea behind ’891 is that deblocking filters should not be applied uniformly across all block edges, but should instead adapt their intensity and scope based on the specific coding history of the adjacent blocks. By recognizing that different coding modes—such as intra-coding versus motion-compensated prediction—introduce different levels of noise and edge reliability, the system can dynamically adjust the filtering strength and the number of pixels modified. This prevents the filter from inadvertently blurring real image features while ensuring that artificial edges created by the compression process are effectively smoothed.

The claims of ’891 focus on an adaptive filtering mechanism that determines the number of pixels to be examined and filtered on each side of a block boundary based on the prediction encoding methods used for the two neighboring blocks. The independent claims specify that the filter receives information identifying whether each block was encoded using intra-coding, copy coding, motion-compensated prediction, or was not coded at all. This metadata is used as a primary parameter to calculate a specific number of pixels to be processed on the first and second sides of the boundary, allowing for asymmetrical filtering if the block types differ.

In practice, the invention implements a multi-step decision process that combines the block type information with the local pixel gradient and the quantization step size. The system calculates an activity parameter to distinguish between a true image edge and a compression artifact. If the difference across the boundary is too large, the filter is disabled to preserve detail. If filtering is warranted, the truncation values derived from the block types (such as Intra or Copy) dictate exactly how many pixels deep into each block the filter will reach, ensuring the smoothing effect is proportional to the likely error in that specific region.

This approach differs from prior solutions by moving away from fixed-window low-pass filters that often over-smooth the image or fail to remove persistent artifacts. By utilizing the coding mode metadata as a control signal for the filter, the invention allows the decoder to make more intelligent decisions about the reliability of the pixel data. This results in a more refined reconstruction where the filtering is aggressive enough to hide block boundaries in flat areas but conservative enough to maintain the sharpness of actual objects within the video frame.