CNN Prediction Based Reversible Data Hiding

Hu, Runwen; Xiang, Shijun

doi:10.1109/lsp.2021.3059202

Cited by 63 publications

(37 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Noting that the reversible watermarking technique can meet the requirement of recovering the cover image and the hiding information in a lossless way [14][15][16], in this paper we adopt the reversible watermarking method proposed in [14] to embed the color palette. Aiming to achieve the requirement of verifying the content integrity in transmission, a practical and convenient approach is to hash the grayscale image and embed the hash value.…”

Section: Reversible Embedding Of the Color Palettementioning

confidence: 99%

Invertible Color-to-Grayscale Conversion by Using Clustering and Reversible Watermarking

Liang

Xiang

2021

2021 IEEE International Conference on Multimedia and Expo (ICME)

Self Cite

View full text Add to dashboard Cite

Invertible color-to-grayscale conversion is a method that embeds the color information into the corresponding grayscale image and extracts the color information to reconstruct the color image when necessary. In this paper, we propose an efficient method by using K-means clustering to generate a color palette and its corresponding grayscale image, and further making use of a reversible watermarking technique to embed the color palette into the grayscale image. For purpose of integrality authentication, the grayscale image is hashed as part of the embedded information before the embedding. In the process of reconstructing the color image, the color palette can be extracted correctly and the grayscale image can be recovered without any loss. Experimental results have shown that the proposed method can provide satisfactory performance.

show abstract

Section: Reversible Embedding Of the Color Palettementioning

confidence: 99%

Invertible Color-to-Grayscale Conversion by Using Clustering and Reversible Watermarking

Liang

Xiang

2021

2021 IEEE International Conference on Multimedia and Expo (ICME)

Self Cite

View full text Add to dashboard Cite

show abstract

“…Several RDH techniques have been proposed for histogram modification (HM) [2,3], difference expansion (DE) [4][5][6], prediction-error expansion (PEE) [7][8][9][10], deep neural networks (DNN) [11,12], and multiple histogram modification (MHM) [3,12,13], and so forth. In recent years, MHM has attracted considerable attention owing to its potential in revealing spatial redundancy in natural images, while DNN suggests a new research direction since DNN serves as a higher performance predictor for predictionbased RDH methods.…”

Section: Introductionmentioning

confidence: 99%

“…DNN-based RDH usually segments a cover image into two layers in a chessboard pattern by employing a deep neural network to predict one layer using the other layer. Hu and Xiang [11] employed a lightweight and computationefficient convolution neural network (CNN) with global optimization for DNN-based data embedding. is CNN predictor was trained using 1000 randomly selected images to prove its effectiveness in increasing the prediction accuracy and improving the embedding performance.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Improving the Security of Reversible Data Hiding Using Multiple Histogram Modification

Chen

Wang

Zhou

et al. 2022

Mathematical Problems in Engineering

View full text Add to dashboard Cite

Reversible data hiding (RDH) allows carrying secret information in cover media without introducing permanent distortion. For a RDH method, the important performance measurements are embedding capacity and image quality. Since embedding capacity is an important requirement in the field of data hiding, it is necessary to consider the security of data embedding in RDH applications. In general, RDH algorithms usually prefer data embedding in simple image regions with low local complexity. As a result, image degradation is alleviated at the cost of poor embedding security. In this study, a novel RDH method is proposed to embed data into complex image regions, wherein the data hiding becomes more secure in defending against modern steganalysis. To measure regional local complexity, the harmonic mean of directional local variances is employed to combine directional pixel differences. To embed data into complex regions instead of smooth regions, multiple histogram modification is adopted and updated for optimized data embedding with higher complexity. Experiment results show that embedding security is significantly improved with a considerable amount of payload and well-preserved image quality.

show abstract

“…However, this suffers from memory or computational constraints, and from overfitting phenomena. Other simple techniques to improve the overall generalization performances of CNNs with little or no overhead include data augmentation [18]- [24] and optimization methods [25]- [28]. These do not require any architectural changes or incur testtime overheads, but may require longer training times.…”

Section: Introductionmentioning

confidence: 99%

Sequential Feature Filtering Classifier

Seo

Lee

Park³

et al. 2021

IEEE Access

View full text Add to dashboard Cite

The Ensemble and mixture of expertise method is the most intuitive and simple way to improve performance in the field of recognition using convolutional neural networks (CNNs). However, Ensemble is difficult to apply in real-time operation applications because the amount of computational overhead and parameters increase in proportion to the number of models. In another, a mixture of expertise that extracts various expertise and combines it is cumbersome to apply because it requires a large change in the network. In this study, we propose Sequential Feature Filtering Classifier (FFC), a simple but effective classifier for CNNs. Using sequential LayerNorm and ReLU, FFC zeroes out low-activation units and preserves high-activation units. The sequential feature filtering process generates multiple features, which are transmitted to a shared classifier, yielding multiple outputs (multiple expertise). FFC can be applied to any CNN with a classifier and it significantly improves the performance with negligible overhead. In this study, the efficacy of FFC is validated extensively on various tasks-ImageNet-1K classification, MS COCO detection, Cityscapes segmentation, and HMDB51 action recognition. Moreover, it is empirically established that FFC can further improve performances using additional techniques, including attention modules.

show abstract

CNN Prediction Based Reversible Data Hiding

Cited by 63 publications

References 24 publications

Invertible Color-to-Grayscale Conversion by Using Clustering and Reversible Watermarking

Invertible Color-to-Grayscale Conversion by Using Clustering and Reversible Watermarking

Improving the Security of Reversible Data Hiding Using Multiple Histogram Modification

Sequential Feature Filtering Classifier

Contact Info

Product

Resources

About