Blind SVD-based watermarking for color images using block-wise quantization

Abstract

The widespread circulation and reuse of digital images makes it increasingly important to verify whether a given image is still consistent with its original content or has undergone substantial modification. Instead of embedding an explicit ownership identifier, this work introduces a blind, imperceptible digital image watermarking scheme that derives a contentdependent perceptual signature directly from the image itself. The watermark is generated from local luminance statistics and embedded into the singular values of block-wise Singular Value Decomposition (SVD) of the RGB channels using quantization. At verification time, the same procedure is applied to a received image, and the extracted watermark is compared to the freshly regenerated one. Consistency between both indicates that the image has not been drastically altered, while significant discrepancies are expected to reveal strong local modifications or severe processing. Unlike traditional ownership watermarking, our scheme acts as a content-bound, signature-like mechanism: a reproducible, image-internal token derived from local block statistics and re-embodied via quantized SVD coefficients. Experimental results show that the proposed scheme maintains high visual quality and can still detect substantial changes under common signal-processing operations, making it suitable as a baseline for semi-fragile, content-bound perceptual signature mechanisms.