Purpose – Reduced difference expansion (RDE) schemes such as Extended RDE (ERDE) are usually evaluated by gross capacity. This ignores the location map and reduction flags that the decoder needs, and every admissible block is used regardless of texture. This study develops an ERDE whose block selection explicitly accounts for that auxiliary information. Design/methods/approach – An Overhead-Aware Adaptive ERDE (OA-ERDE) orders 2 × 2 blocks by a texture measure computed from never-modified base pixels, chooses per block between plain and reduced expansion, and minimises estimated distortion under a net-capacity constraint that uses the actually encoded auxiliary data. OA-ERDE, RDE, ERDE, adaptive ERDE and a prediction-error expansion (PEE) benchmark were implemented in Python and compared on 32 grayscale images with paired tests in IBM SPSS Statistics 25. Findings – Mean maximum net capacity rose from 0.199 bpp (ERDE) and 0.130 bpp (RDE) to 0.660 bpp (173,119 bits). At 20 kbit, OA-ERDE improved PSNR by 10.70, 7.81 and 9.32 dB over ERDE, RDE and adaptive ERDE (Holm-adjusted p < 0.001) but remained 3.68 dB below PEE. All 1,615 successful embeddings were recovered bit-exactly. Research implications/limitations – Results are limited to 512 × 512 grayscale natural images and single-layer embedding. OA-ERDE embedding took 4.17 s per image, about 23 times longer than PEE, whereas decoding took 0.017 s. Originality/value – Counting auxiliary information inside the selection objective, rather than after embedding, changes both which blocks are used and how they are expanded. The study specifies an ERDE implementation with a defined reduction-and-recovery procedure and demonstrates exact image reconstruction across the reported successful embedding trials.
Copyrights © 2026