Spatial Domain Watermarking Scheme for Colored Images Based on Log-average Luminance

In this paper a new watermarking scheme is presented based on log-average luminance. A colored-image is divided into blocks after converting the RGB colored image to YCbCr color space. A monochrome image of 1024 bytes is used as the watermark. To embed the watermark, 16 blocks of size 8X8 are selected and used to embed the watermark image into the original image. The selected blocks are chosen spirally (beginning form the center of the image) among the blocks that have log-average luminance higher than or equal the log-average luminance of the entire image. Each byte of the monochrome watermark is added by updating a luminance value of a pixel of the image. If the byte of the watermark image represented white color (255) a value is added to the image pixel luminance value, if it is black (0) the is subtracted from the luminance value. To extract the watermark, the selected blocks are chosen as the above, if the difference between the luminance value of the watermarked image pixel and the original image pixel is greater than 0, the watermark pixel is supposed to be white, otherwise it supposed to be black. Experimental results show that the proposed scheme is efficient against changing the watermarked image to grayscale, image cropping, and JPEG compression.

💡 Research Summary

The paper proposes a spatial‑domain watermarking method for color images that leverages log‑average luminance to select embedding locations. After converting an RGB image to the YCbCr color space, the authors compute the log‑average luminance of the entire image. The image is then partitioned into 8 × 8 blocks, and only those blocks whose own log‑average luminance is greater than or equal to the global value are considered candidates. From this candidate set, 16 blocks are chosen in a spiral order starting from the image centre, a strategy intended to concentrate the watermark in visually important regions while providing resilience against cropping.

The watermark itself is a 1024‑byte monochrome bitmap (8192 bits). Embedding is performed by modifying the luminance (Y) component of a single pixel in each selected block. If a watermark bit corresponds to white (value 255), a constant α is added to the pixel’s luminance; if the bit is black (value 0), α is subtracted. The value of α is selected experimentally to balance imperceptibility against robustness: a larger α improves detection after attacks but may introduce visible artifacts.

Extraction requires both the original (unwatermarked) image and the watermarked version. The same block‑selection algorithm is applied to locate the modified pixels, and the difference in luminance is examined. A positive difference indicates a white watermark bit, while a non‑positive difference indicates black. This differential approach yields high detection accuracy when the original image is available.

The authors evaluate robustness against three common manipulations: conversion to grayscale, cropping, and JPEG compression. Grayscale conversion removes the chroma channels (Cb, Cr) but leaves the Y channel untouched, so the watermark survives with minimal loss. Cropping tests show that because the embedding blocks are centrally located, even when peripheral blocks are removed, enough embedded pixels remain to recover a substantial portion of the watermark. JPEG compression experiments, conducted over a range of quality factors, demonstrate that with an appropriately chosen α the watermark survives typical lossy compression levels, as the luminance modifications exceed the quantization noise introduced by JPEG.

While the method achieves low computational complexity and reasonable robustness, it has notable limitations. First, the need for the original image during extraction restricts its applicability to scenarios where the host image can be securely stored or transmitted alongside the watermarked version. Second, the fixed α does not adapt to local image characteristics; a globally optimal α may be sub‑optimal for regions with very high or very low contrast, potentially causing either detection failures or visible distortion. Third, the spiral block‑selection rule is static and does not consider image content such as edges or texture, which could be exploited to improve imperceptibility and robustness.

Future work suggested by the authors includes developing a blind extraction scheme (i.e., without the original image), implementing adaptive α selection based on local statistics, and exploring content‑aware block selection strategies. Such enhancements could raise the method’s security level, broaden its practical deployment, and align it with modern requirements for robust, invisible digital watermarking in multimedia distribution.

In summary, the paper introduces a novel, luminance‑centric spatial watermarking framework that combines log‑average luminance‑based block selection with simple additive/subtractive embedding. The experimental results confirm that the scheme tolerates grayscale conversion, moderate cropping, and JPEG compression while maintaining visual quality, thereby offering a lightweight yet effective solution for copyright protection and integrity verification of color images.