RBP-BDE: Mitigating Error Accumulation in Bit-Depth Enhancement Through Reverse Bit-Plane Prediction

In the field of visual Internet of Things (VIoT), image bit-depth enhancement (BDE) technology faces dual challenges of false contour artifacts and limited computational resources. Although deep enhancement methods based on bit-plane decomposition can alleviate this problem, existing plane-by-plane approaches suffer from two major limitations: a mismatch between error contributions and network capabilities, as well as cross-plane error accumulation. To address these issues, this paper proposes a new plane-by-plane enhancement network named RBP-BDE. First, guided by the Universal Approximation Theorem, network parameters are allocated based on bit-plane sensitivity to prioritize processing of more significant and error-sensitive planes. Second, we design a residual recursive correction module to suppress error propagation by injecting original image information across layers. Finally, a state-space-driven feature interaction module is adopted to model long-range dependencies across bit-planes through implicit state differential equations. Experiments demonstrate that our method outperforms previous BDE methods on both natural and animated images. Visual comparisons confirm that the high bit-depth (HBD) images generated by the proposed algorithm exhibit fewer false contours and clearer, more accurate details.