Subjective demonstrations of the Spectrally Adaptive Noise Filling Tool in a HEVC-like still-image codec
The following illustrations serve as a demonstration of the subjective effect of using SANFT in a transform-based still image codec like HEVC. The underlying codec chosen for this demonstration is HEVC [1] with an additional quad-tree/binary-tree (QTBT) tool and extended maximum transform unit size of 64×64 luminance pels [2]. Note that the coding tree unit size remains 64×64 pels, as specified in the HEVC standard [1].
The presented images were transcoded, with visual transparency, to high-bit-rate JPEG in order to limit the download durations for the viewers.
BQTerrace, uncoded input (HD, 1920×1080, lossless size: 5480 KB)
BQTerrace, HEVC+QTBT without SANFT, QP 30 (HD, 1920×1080, coded size: 135 KB)
BQTerrace, HEVC+QTBT with SANFT 1.1, QP 30 (HD, 1920×1080, coded size: 137 KB)
HomelessSleeping, uncoded input (UHD, 2560×1600, lossless size: 9770 KB)
HomelessSleeping, HEVC+QTBT without SANFT, QP 30 (UHD, 2560×1600, coded size: 30 KB)
HomelessSleeping, HEVC+QTBT with SANFT 1.1, QP 30 (UHD, 2560×1600, coded size: 31 KB)
ParkScene, uncoded input (HD, 1920×1080, lossless size: 5415 KB)
ParkScene, HEVC+QTBT without SANFT, QP 30 (HD, 1920×1080, coded size: 91 KB)
ParkScene, HEVC+QTBT with SANFT 1.1, QP 30 (HD, 1920×1080, coded size: 93 KB)