Image Watermark Embedding Using Color-Channel Frequency Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Watermarking technologies based on deep learning face performance deterioration and image quality degradation when detecting watermarks in images not used in training, affecting their effectiveness in unauthorized use prevention.
Innovation Solution
An electronic apparatus identifies a color channel with the highest complexity by analyzing frequency components and embedding a watermark with varying strength and location-based information, using discrete cosine transform coefficients to optimize watermarking in images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a watermarking technology based on deep learning uses a trained artificial intelligence model to embed and extract watermarks, then the watermark detection performance is improved for images similar to training images, but the performance deteriorates and image quality degrades for other images
Solution Approach 1:
The patent divides the image into multiple frequency bands (low, medium, high frequencies) and applies different watermark embedding strategies to each band. Specifically, watermarks are embedded with different strengths and methods in different frequency regions, allowing the system to optimize for both detection accuracy and generalizability across diverse images without relying solely on training-specific patterns
2Reliability
If the embedding strength of the watermark is increased to improve detectability, then the watermark strength is improved, but the image quality degradation increases
Solution Approach 1:
The patent embeds watermarks with varying strengths in different frequency bands rather than using uniform embedding strength across the entire image. Low-frequency regions receive different embedding treatment compared to high-frequency regions, thereby maintaining watermark detectability while minimizing visual degradation and preserving image quality
Solution Approach 2:
The patent transforms the image into the frequency domain using discrete cosine transform (DCT) and embeds watermarks in the frequency coefficients rather than directly in spatial pixel values. This parameter transformation allows watermark embedding that is less perceptible to human vision while maintaining robustness, effectively decoupling watermark strength from visible image quality degradation
Data Source
AI summary
An electronic apparatus includes: at least one processor including processing circuitry; and memory storing instructions, wherein the instructions, when executed by the at least one processor individually or collectively, cause the electronic apparatus to: obtain information on sizes of frequency components for a plurality of color channels constituting an image, identify a color channel having a size ratio of a high frequency component that is highest among the plurality of color channels, based on a size of the high frequency component of the identified color channel for each of a plurality of areas constituting the image, identify an embedding strength of a watermark for each of the plurality of areas, embed, into the watermark, (i) information corresponding to the embedding strength for each of the plurality of areas and (ii) locations of each of the plurality of areas, and embed, into the image, the watermark.