PROJECT TITLE :

Deep Reconstruction of Least Significant Bits for Bit-Depth Expansion

ABSTRACT:

To display a low bit-depth image on a high-bit-depth panel, a technique called bit-depth expansion is necessary. To fill in the missing least significant bits, current BDE algorithms generally rely on traditional approaches and suffer from a wide range of perceptible artefacts. The BDE method presented in this paper employs a deep residual network-based approach. Two pathways for reconstructing flat and non-flat environments have been proposed based on the differences in properties between the two. Flat-area-channel preprocessing, on the other hand, employs a basic yet effective local adaptive adjustment preprocessing. The proposed method can improve the flat area's subjective quality even more by combining the advantages of both the standard debanding methods and network-based reconstruction. The proposed BDE network has been tested on a variety of image sets and has shown promising visual and quantitative results.