Video summarization via block sparse dictionary selection

Ma, Mingyang; Mei, Shaohui; Wan, Shuai; Hou, Junhui; Wang, Zhiyong; Feng, David Dagan

doi:10.1016/j.neucom.2019.07.108

Cited by 71 publications

(32 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Euclidean distance and Mahalanobis distance are usually used to measure the distance between different features. And the Euclidean distance D e and Mahalanobis distance D m are computed by (2) and 3, respectively:…”

Section: B Metric Learningmentioning

confidence: 99%

See 1 more Smart Citation

Rotation-Invariant Feature Learning for Object Detection in VHR Optical Remote Sensing Images by Double-Net

et al. 2020

Self Cite

View full text Add to dashboard Cite

Rotation-invariant feature extraction is crucial for object detection in Very High Resolution (VHR) optical remote sensing images. Although convolutional neural networks (CNNs) are good at extracting the translation-invariant features and have been widely applied in computer vision, it is still a challenging problem for CNNs to extract rotation-invariant features in VHR optical remote sensing images. In this paper we present a novel Double-Net with sample pairs from the same class as inputs to improve the performance of object detection and classification in VHR optical remote sensing images. Specifically, the proposed Double-Net contains multiple channels of CNNs in which each channel refers to a specific rotation direction and all CNNs share identical weights. Based on the output features of all channels, multiple instance learning algorithm is employed to extract the final rotation-invariant features. Experimental results on two publicly available benchmark datasets, namely Mnist-rot-12K and NWPU VHR-10, demonstrate that the presented Double-Net outperforms existing approaches on the performance of rotation-invariant feature extraction, which is especially effective under the situation of small training samples. INDEX TERMS Convolution neural network (CNN), feature learning, object detection, rotation-invariant, very high resolution (VHR).

show abstract

Section: B Metric Learningmentioning

confidence: 99%

“…To address the rotation-invariant problem, most of conventional machine learning approaches employ handcrafted or shallow-learning-based features [2], [3]. One of the The associate editor coordinating the review of this manuscript and approving it for publication was Rajeeb Dey .…”

Section: Introductionmentioning

confidence: 99%

Rotation-Invariant Feature Learning for Object Detection in VHR Optical Remote Sensing Images by Double-Net

et al. 2020

Self Cite

View full text Add to dashboard Cite

show abstract

“…Therefore, we need to design a set of high-security encryption scheme to ensure that digital images in the mobile Internet transmission are secure and safe. However, different from the text files, digital images have some intrinsic features, such as bulk data capacity, high redundancy and strong correlation [1], [2], [56], [57], which make traditional encryption algorithms, such as data encryption standard, advanced encryption standard, and triple data encryption algorithm, are unsuitable for the digital images. With the development of chaos theory, researchers begin to realize that the chaotic system could be used for digital image encryption because of its extreme sensitive to initial…”

Section: Introductionmentioning

confidence: 99%

“…In recent years, a lot of digital image encryption algorithms are designed based on the chaotic system [3]- [7], [32], [41], [42], [51]- [54], [63]- [65]. Some other new technologies, such as the DNA encoding technology [8]- [10], [50], [67], the simulated annealing algorithm [11], the cellular automata [2], [12], [29], the affine transformation [13], [38], the compressive sensing [39], [40] and the S-box [14], [15], [30], [58]- [60] are proposed combining with chaos. Nowadays, image encryption becomes a hot topic for researchers.…”

Section: Introductionmentioning

confidence: 99%

A Plaintext-Related Dynamical Image Encryption Algorithm Based on Permutation-Combination-Diffusion Architecture

Liu

Jiang

et al. 2020

IEEE Access

View full text Add to dashboard Cite

In this paper, a novel high security plaintext-related dynamical image encryption algorithm is proposed. Compared with the classical permutation-diffusion encryption architecture, we add the combination operation before the diffusion, which uses the 3D intertwining Logistic map (3D-ILM) to generate dynamic interference matrix with the same size of the grayscale plaintext image, and then combine the permutated image with interference matrix. It can be seen as an improvement to enhance the connection between permutation and diffusion operation, which makes adversaries are difficult to crack the permutation and diffusion algorithm separately. Besides, the permutated pixels are hidden in the dynamic interference matrix during the combination process, which reduces the correlation among local adjacent pixels. Thus, it can increase the resistance of chosen-plaintext attack. After the permutation, combination, and diffusion operation, a dynamic color cipher image is generated. Note that the generated color cipher image is dynamic changing even the same plaintext image is encrypted with identical secret keys. Ultimately, simulation results and security analysis show that the proposed algorithm has high-security performance and can resist most known common attacks.

show abstract

“…According to CS theory, the sampling rates (srs) for sparse or compressible signals can be far lower than Nyquist srs. Sparse representation have been widely used in image and video processing [5][6][7]. The application of CS in the hyperspectral imaging field has also received wide attention.…”

Section: Introductionmentioning

confidence: 99%

Spatial-Spectral Joint Compressed Sensing for Hyperspectral Images

Wang

Hua

et al. 2020

IEEE Access

View full text Add to dashboard Cite

Compressed sensing is one of the key technologies to reduce the volume of hyperspectral image for real-time storage and transmission. Reconstruction based on spectral unmixing show tremendous potential in hyperspectral compressed sensing compared with other conventional algorithms that directly reconstruct images. In this paper, a joint spatial-spectral joint compressed sensing scheme is proposed. In this scheme, compressed hyperspectral data are collected by spatial-spectral hybrid compressed sampling. As for the reconstruction, an objective function is developed by introducing the fidelity constraints of spatial and spectral measurements and the row sparsity constraint of abundance that guarantee the precise reconstruction of hyperspectral images and obtain endmembers and abundances as by-products accurately. An augmented Lagrangian type algorithm is meticulously elaborated to solve the above optimization problem. Extensive experimental results on several real hyperspectral datasets indicate that the proposed approach can achieve a reconstruction accuracy higher than those of other state-of-the-art methods. The efficiency and feasibility of the proposed scheme give it great potential in hyperspectral compressed sensing.

show abstract

Video summarization via block sparse dictionary selection

Cited by 71 publications

References 31 publications

Rotation-Invariant Feature Learning for Object Detection in VHR Optical Remote Sensing Images by Double-Net

Rotation-Invariant Feature Learning for Object Detection in VHR Optical Remote Sensing Images by Double-Net

A Plaintext-Related Dynamical Image Encryption Algorithm Based on Permutation-Combination-Diffusion Architecture

Spatial-Spectral Joint Compressed Sensing for Hyperspectral Images

Contact Info

Product

Resources

About