Wenbo Wan

Video hashing based on appearance and attention features fusion via DBN

Abstract Video hashing has attracted increasing attention in the field of large-scale video retrieval. However, only low-level features or their combinations, referred to as appearance features, are used to generate the video hash in most of the existing video hashing algorithms and these kinds of features are referred to as appearance features. In this paper, a visual attention model is used to extract visual attention features, and the video hash is generated from a fusion of visual-appearance and visual-attention features via a deep belief network (DBN) to obtain representative video features. In addition, hash distance is taken as a vector to measure the similarity between hashes. BER is used as the amplitude of hash distance and the vector cosine similarity is used as the angle of hash distance. Experimental results demonstrate that the fusion of visual appearance and attention features brings about better performance of video hash on recall and precision rates, and the angle of hash distance is useful to improve the accuracy of hash matching.

Logarithmic Spread-Transform Dither Modulation watermarking Based on Perceptual Model

Logarithmic Quantization Index Modulation (LQIM) is an important extension of the original quantization-based watermarking method. However, it is well known that it is sensitive to valumetric scaling attack and easy to result in sign error after quantization and attacks. For that, in this paper, we propose a new method, namely Logarithmic Spread-Transform Dither Modulation Based on Perceptual Model (LSTDM-WM). In this regard the host signal is first projected onto a random vector and transformed using a novel Logarithmic Quantization function. Then the transformed signal is quantized regarding the watermark data and the watermarked signal is obtained by applying inverse transform to the quantized signal. The perceptual model is further exploited to adjust the quantization step adaptively for watermark embedding. Experimental results indicate that our proposed scheme overcomes two challenges cited above and has superior performance in comparison with conventional LQIM and former proposed schemes of STDM.

A two-stage learning approach to face recognition

Abstract This paper introduces the Collaborative Representation (CR) techniques to small sample size conditions, and propose a Two-Stage learning approach to face recognition based on Collaborative Representation (TSCR). Based on the assumption that the same class samples should lie in the same subspace, we first use the unlabeled samples as dictionary atoms to construct each labeled sample, and obtain the collaborative coefficients by CR. The unlabeled sample with the largest collaborative coefficient is assigned the same class label as the reconstructed labeled sample, and is added to the labeled data set. This process is repeated until about half of the unlabeled samples are labeled and added to the labeled dataset. After that, we employ the original CR approach to classify the left unlabeled samples based on the newly labeled dataset. Experimental results demonstrate that the proposed TSCR is effective on face recognition.

Improved logarithmic spread transform dither modulation using a robust perceptual model

In the quantization-based watermarking framework, the perceptual just noticeable distortion (JND) model has been widely used to determine the quantization step size, as it can be used for the better tradeoff between imperceptibility and robustness. However, the calculated JND values will change as watermark embedding can affect the texture and luminance of the image. Consequently, the changes of JND values will lead to watermark-extraction errors. In this paper, the authors present an improved logarithmic spread transform dither modulation (STDM) watermarking approach using a best-matched DCT-based perceptual JND model, which can be insensitive to the changes caused by watermark embedding and attacks. Experimental results confirm the improved robustness performance of the JND model in the watermarking framework. Simulation results show that the proposed scheme is more robust than the existing JND model-based watermarking algorithms with the uniform fidelity, and our proposed scheme has a superior performance compared with the former proposed perceptual STDM schemes.

Semi-supervised modality-dependent cross-media retrieval

In this paper, we propose a modality-dependent cross-media retrieval approach under semi-supervised conditions. The approach utilizes both labeled samples and unlabeled ones to obtain two couples of projection matrices and uses feature distance to represent the semantic information of unlabeled samples in the optimization process, so as to fully utilize the data structural information. Different from supervised modality-dependent cross-media retrieval approaches which use labeled samples and fixed semantic information, the proposed approach makes full use of the global data distribution property and the semantic information of both labeled and unlabeled samples. Experiments on benchmark datasets show its superiority over the compared methods.

Joint graph regularization based modality-dependent cross-media retrieval

Cross-media retrieval returns heterogeneous multimedia data of the same semantics for a query object, and the key problem for cross-media retrieval is how to deal with the correlations of heterogeneous multimedia data. Many works focus on mapping different modal data into an isomorphic space, so the similarities between different modal data can be measured. Inspired by this idea, we propose a joint graph regularization based modality-dependent cross-media retrieval approach (JGRMDCR), which takes into account the one-to-one correspondence between different modal data pairs, the inter-modality similarities and the intra-modality similarities. Meanwhile, according to the modality of the query object, this method learns different projection matrices for different retrieval tasks. Experimental results on benchmark datasets show that the proposed approach outperforms the other state-of-the-art algorithms.

A Novel STDM Watermarking Using Visual Saliency-Based JND Model

The just noticeable distortion (JND) model plays an important role in measuring the visual visibility for spread transform dither modulation (STDM) watermarking. However, the existing JND model characterizes the suprathreshold distortions with an equal saliency level. Visual saliency (VS) has been widely studied by psychologists and computer scientists during the last decade, where the distortions are more likely to be noticeable to any viewer. With this consideration, we proposed a novel STDM watermarking method for a monochrome image by exploiting a visual saliency-based JND model. In our proposed JND model, a simple VS model is employed as a feature to reflect the importance of a local region and compute the final JND map. Extensive experiments performed on the classic image databases demonstrate that the proposed watermarking scheme works better in terms of the robustness than other related methods.

Improved Spread Transform Dither Modulation Using Luminance-Based JND Model

In the quantization-based watermarking framework, perceptual just noticeable distortion (JND) model has been widely used to determine the quantization step size to provide a better tradeoff between fidelity and robustness. However, the perceptual parameters computed in the embedding procedure and the detecting procedure are different, as the image has been altered by watermark embedding. In this paper, we incorporate a new DCT-based perceptual JND model, which not only shows better consistency with the HVS characteristics compared to the conventional models, but also can be invariant to the changes in the watermark framework. Furthermore, an improved spread transform dither modulation (STDM) watermarking scheme based on the new JND model is proposed. Experimental results show that the proposed scheme provides powerful resistance against common attacks, especially in robustness against Gauss noise, amplitude scaling and JPEG compression.

An improved FRUC scheme based on motion vector refinement

In this paper, a low-complexity and high-efficiency scheme is developed for frame rate up conversion (FRUC). Texture based adaptive motion estimation is carried out for improving the accuracy of motion vectors. In addition, the proposed scheme comprises a motion vector post-processing method which corrects the outliers in nine directions. Moreover, in order to deal with the artifacts caused by the problem of overlaps and holes, a simple motion refinement method by calculating motion vectors of overlaps and holes is proposed. Experimental results verify the superiority of our work in both objective and subjective performances compared with conventional methods.

View-invariant gait recognition based on kinect skeleton feature

Gait recognition is a popular remote biometric identification technology. Its robustness against view variation is one of the challenges in the field of gait recognition. In this paper, the second-generation Kinect (2G–Kinect) is used as a tool to build a 3D–skeleton-based gait dataset, which includes both 2D silhouette images captured by 2G–Kinect and their corresponding 3D coordinates of skeleton joints. Given this dataset, a human walking model is constructed. Referring to the walking model, the length of some specific skeletons is selected as the static features, and the angles of swing limbs as the dynamic features, which are verified to be view-invariant. In addition, the gait recognition abilities of the static and dynamic features are investigated respectively. Given the investigation, a view-invariant gait recognition scheme is proposed based on the matching-level-fusion of the static and dynamic features, and the nearest neighbor (NN) method is used for recognition. Comparison between the existing Kinect-based gait recognition method and the proposed one on different datasets show that the proposed one has better recognition performance.