Perceptual Hashing With Complementary Color Wavelet Transform and Compressed Sensing for Reduced-Reference Image Quality Assessment

Mengzhu Yu, Zhenjun Tang, Xianquan Zhang, Bineng Zhong, Xinpeng Zhang

Abstract

Ensuring image quality during transmission and processing is a critical challenge in modern data systems. This paper proposes a novel Reduced-Reference Image Quality Assessment (RR IQA) method based on perceptual hashing. By integrating a Complementary Color Wavelet Transform (CCWT) and block-based Compressed Sensing (CS), the algorithm generates a compact 720-bit hash that captures both structural and color information. Experimental results demonstrate that this approach effectively balances robustness to content-preserving edits with high sensitivity to distortions, achieving a 0.9006 correlation with human subjective scores on the LIVE database.

1. Introduction

In the era of big data, Image Quality Assessment (IQA) is essential for maintaining user experience across various platforms. IQA methods are generally categorized into Full-Reference (FR), No-Reference (NR), and Reduced-Reference (RR). While FR methods are highly accurate, they require the full original image, consuming significant bandwidth. NR methods are bandwidth-efficient but often lack accuracy. RR IQA offers a middle ground by using a compact "summary" or hash of the original image.

The primary challenge in RR IQA is the trade-off between robustness (ignoring minor, non-destructive edits like resizing) and sensitivity (detecting quality-degrading artifacts like blur or noise). Many existing methods struggle by ignoring color information or failing to capture subtle structural changes.

2. Proposed Methodology

The authors introduce a multi-stage pipeline designed to create a perceptually relevant hash that accounts for human visual system characteristics.

Flowchart of the proposed perceptual hashing algorithm pipeline.

2.1 Complementary Color Wavelet Transform (CCWT)

Standard wavelets often focus on luminance, discarding color data. In contrast, CCWT decomposes images into four operators—White-Black, Red-Cyan, Green-Magenta, and Blue-Yellow. This ensures that distortions affecting specific color channels are preserved in the hash.

2.2 Multi-Directional Edge Detection and Feature Extraction

An 8-direction Sobel operator is applied to the color sub-bands to capture fine-grained structural details. These features are then processed using Block-Based Compressed Sensing (CS). While CS is typically used for image reconstruction, it is used here for feature extraction. The process is naturally robust to global edits but highly sensitive to pixel-level distortions that disrupt the signal's sparsity.

2.3 Hash Construction

The extracted features are quantized into a compact sequence of 80 integers (720 bits). Image comparison is performed by calculating the Euclidean distance (L2 norm) between these strings, providing a fast and efficient quality metric.

3. Experimental Results and Key Insights

The algorithm was evaluated using four major databases: UCID (for robustness), COCO 2017 (for discrimination), and LIVE/TID2013 (for quality assessment).

3.1 Performance Metrics

The proposed method achieved an Area Under the Curve (AUC) of 0.99986 for classification accuracy. On the LIVE database, its quality predictions showed a 0.9006 correlation with human Mean Opinion Scores (MOS), notably outperforming the classic Full-Reference metric, PSNR (0.8756).

3.2 Sensitivity to Signal Sparsity

A key insight of this research is the use of CS as a distortion filter. Because CS relies on signal sparsity, distortions like white noise significantly alter the resulting hash, making them easily detectable.

An example of a noisy image; such distortions break signal sparsity, making them highly detectable by the CS-based hash.

3.3 Efficiency and Versatility

The algorithm is highly efficient, taking approximately 0.22 seconds to generate a hash. In the TID2013 database, which includes 24 distortion types, the method ranked first among RR algorithms in 17 categories, demonstrating its versatility across different types of image degradation.

4. Conclusion

The proposed perceptual hashing algorithm provides a robust and sensitive solution for RR IQA. By integrating color-aware wavelet transforms with compressed sensing, it achieves performance comparable to or better than some full-reference metrics while maintaining a very small data footprint. Future work may involve extending this multi-directional approach to video sequences to address temporal quality assessment.

Summary updated

Abstract

Image quality assessment (IQA) is an important task of image processing and has diverse applications, such as image super-resolution reconstruction, image transmission and monitoring systems. This paper proposes a perceptual hashing algorithm with complementary color wavelet transform (CCWT) and compressed sensing (CS) for reduced-reference (RR) IQA. The CCWT is exploited to decompose input color image into different sub-bands. Since the calculation of CCWT uses all color channels without discarding any information, the distortions introduced by digital operations on color channels are preserved in the CCWT sub-bands. The block-based CS is used to extract features from the CCWT sub-bands. As the Euclidean distance between the block-based CS features is slightly influenced by content-preserving operations, perceptual features constructed by Euclidean distances are robust, discriminative and compact. Hash sequence is finally determined by quantifying the perceptual features. Effectiveness of the proposed hashing is verified by various experiments on four open image databases. Experimental results demonstrate that the proposed hashing is superior to some state-of-the-art algorithms in terms of classification and RR IQA application.
BibTeX
@article{yu2022perceptual,
  author = {Mengzhu Yu and Zhenjun Tang and Xianquan Zhang and Bineng Zhong and Xinpeng Zhang},
  title = {{Perceptual Hashing With Complementary Color Wavelet Transform and Compressed Sensing for Reduced-Reference Image Quality Assessment}},
  journal = {IEEE Trans. Circuits Syst. Video Technol.},
  year = {2022},
  doi = {10.1109/tcsvt.2022.3190273},
}
Figure