This technology is an image processing method that calculates a reference value for pixel vectors and an activity level representing local variation in DCT-encoded images. It classifies images into three models—LAM, MAM, and HAM—based on this activity level, then selectively applies a flat-edge filter or a model-specific filter. If the reference value is below a certain threshold, the image is considered free of blocking artifacts, and filtering is skipped.
Block-based DCT compression, such as JPEG, MPEG, and H.26x, often results in blocking artifacts where block boundaries become visible because inter-block correlations are not considered. Conventional low-pass filtering post-processing often causes excessive blurring that degrades object edges, while statistical model or POCS-based methods are too computationally intensive for real-time video processing.
This technology sets differential filter application ranges for each model based on a first threshold determined by quantization factors and pixel vector standard deviation, applying appropriate deblocking intensity to flat and complex regions. This reduces blocking artifacts with low computational complexity while preserving object edges and sharpness. It can be applied to video codec post-processing, streaming players, and digital broadcast receivers to improve perceived quality in real-time, even for low-quality, high-compression videos.
N/A
