The residue number system (RNS) is a method for representing an integer x as an n-tuple of its residues with respect to a given set of moduli. In RNS, addition, subtraction, and multiplication can be carried out by independent operations with respect to each modulus. Therefore, an n-fold speedup can be achieved by parallel processing. The main disadvantage of RNS is that we cannot efficiently compare the magnitude of two integers or determine the sign of an integer. Two general methods of comparison are to transform a number in RNS to a mixed-radix system or to a radix representation using the Chinese remainder theorem (CRT). We used the CRT to derive an equation approximating a value of x relative to M, the product of moduli. Then, we propose two algorithms that efficiently evaluate the equation and output a sign bit. The expected number of steps of these algorithms is of order n. The algorithms use a lookup table that is (n+3) times as large as M, which is reasonably small for most applications including cryptography.

With the rapid development of multimedia, violent video can be easily accessed in games, movies, websites, and so on. Identifying violent videos and rating violence extent is of great importance to media filtering and children protection. Many previous studies only address the problems of violence scene detection and violent action recognition, yet violence rating problem is still not solved. In this paper, we present a novel video-level rating prediction method to estimate violence extent automatically. It has two main characteristics: (1) a two-stream network is fine-tuned to construct effective representations of violent videos; (2) a violence rating prediction machine is designed to learn the strength relationship among different videos. Furthermore, we present a novel violent video dataset with a total of 1,930 human-involved violent videos designed for violence rating analysis. Each video is annotated with 6 fine-grained objective attributes, which are considered to be closely related to violence extent. The ground-truth of violence rating is given by pairwise comparison method. The dataset is evaluated in both stability and convergence. Experiment results on this dataset demonstrate the effectiveness of our method compared with the state-of-art classification methods.

Deep neural networks have achieved great success in visual tracking by learning a generic representation and leveraging large amounts of training data to improve performance. Most generic object trackers are trained from scratch online and do not benefit from a large number of videos available for offline training. We present a real-time generic object tracker capable of incorporating temporal information into its model, learning from many examples offline and quickly updating online. During the training process, the pre-trained weight of convolution layer is updated lagging behind, and the input video sequence length is gradually increased for fast convergence. Furthermore, only the hidden states in recurrent network are updated to guarantee the real-time tracking speed. The experimental results show that the proposed tracking method is capable of tracking objects at 150 fps with higher predicting overlap rate, and achieves more robustness in multiple benchmarks than state-of-the-art performance.

We present an improved constant-round secure two-party protocol for integer comparison functionality, which is one of the most fundamental building blocks in secure computation. Our protocol is in the so-called client-server model, which is utilized in real-world MPC products such as Sharemind, where any number of clients can create shares of their input and distribute to the servers who then jointly compute over the shares and return the shares of the result to the client. In the client-aided client-server model, as mentioned briefly by Mohassel and Zhang (S&P'17), a client further generates and distributes some necessary correlated randomness to servers. Such correlated randomness admits efficient protocols since otherwise, servers have to jointly generate randomness by themselves, which can be inefficient. In this paper, we improve the state-of-the-art constant-round comparison protocols by Damgå rd et al. (TCC'06) and Nishide and Ohta (PKC'07) in the client-aided model. Our techniques include identifying correlated randomness in these comparison protocols. Along the way, we also use tree-based techniques for a building block, which deviate from the above two works. Our proposed protocol requires only 5 communication rounds, regardless of the bit length of inputs. This is at least 5 times fewer rounds than existing protocols. We implement our secure comparison protocol in C++. Our experimental results show that this low-round complexity benefits in high-latency networks such as WAN. We also present secure Min/Argmin protocols using the secure comparison protocol.

Secure two-party comparison plays a crucial role in many privacy-preserving applications, such as privacy-preserving data mining and machine learning. In particular, the available comparison protocols with the appropriate input/output configuration have a significant impact on the performance of these applications. In this paper, we firstly describe a taxonomy of secure two-party comparison protocols which allows us to describe the different configurations used for these protocols in a systematic manner. This taxonomy leads to a total of 216 types of comparison protocols. We then describe conversions among these types. While these conversions are based on known techniques and have explicitly or implicitly been considered previously, we show that a combination of these conversion techniques can be used to convert a perhaps less-known two-party comparison protocol by Nergiz et al. (IEEE SocialCom 2010) into a very efficient protocol in a configuration where the two parties hold shares of the values being compared, and obtain a share of the comparison result. This setting is often used in multi-party computation protocols, and hence in many privacy-preserving applications as well. We furthermore implement the protocol and measure its performance. Our measurement suggests that the protocol outperforms the previously proposed protocols for this input/output configuration, when off-line pre-computation is not permitted.

Establishing local visual correspondences between images taken under different conditions is an important and challenging task in computer vision. A common solution for this task is detecting keypoints in images and then matching the keypoints with a feature descriptor. This paper proposes a robust and low-dimensional local feature descriptor named Adaptively Integrated Gradient and Intensity Feature (AIGIF). The proposed AIGIF descriptor partitions the support region surrounding each keypoint into sub-regions, and classifies the sub-regions into two categories: edge-dominated ones and smoothness-dominated ones. For edge-dominated sub-regions, gradient magnitude and orientation features are extracted; for smoothness-dominated sub-regions, intensity feature is extracted. The gradient and intensity features are integrated to generate the descriptor. Experiments on image matching were conducted to evaluate performances of the proposed AIGIF. Compared with SIFT, the proposed AIGIF achieves 75% reduction of feature dimension (from 128 bytes to 32 bytes); compared with SURF, the proposed AIGIF achieves 87.5% reduction of feature dimension (from 256 bytes to 32 bytes); compared with the state-of-the-art ORB descriptor which has the same feature dimension with AIGIF, AIGIF achieves higher accuracy and robustness. In summary, the AIGIF combines the advantages of gradient feature and intensity feature, and achieves relatively high accuracy and robustness with low feature dimension.

This paper deals with a novel, interesting problem of detecting frauds in comparison-shopping services (CSS). In CSS, there exist frauds who perform excessive clicks on a target item. They aim at making the item look very popular and subsequently ranked high in the search and recommendation results. As a result, frauds may distort the quality of recommendations and searches. We propose an approach of detecting such frauds by analyzing click behaviors of users in CSS. We evaluate the effectiveness of the proposed approach on a real-world clickstream dataset.

A lot of encryption and watermarking schemes have been developed as countermeasures to protect copyrights of broadcast or multicast content from malicious subscribers (traitors) that make pirate receivers (PRs) to use the content illegally. However, solo use of these schemes does not necessarily work well. Traitor tracing encryption schemes are a type of broadcasting encryption and have been developed for broadcasting and multicast services. There are multiple distinct decryption keys for each encryption key, and each service subscriber is given a unique decryption key. Any subscriber that redistributes his or her decryption key to a third party or who uses it and maybe other keys to make a PR can be identified with using the tracing algorithm of the scheme that is used by the services. However, almost all previous schemes have the same weakness; that is, they are vulnerable to an attack (content comparison attack). This is a concrete example such that solo use of the scheme does not work well. The attack involves multiple distinct decryption keys and a content-data comparison mechanism. We have developed a method, called complementary traitor tracing method (CTT), that makes traitor tracing schemes secure against content comparison attacks. It makes it impossible for PRs to distinguish ordinary content data from test data and makes traitor tracing schemes effective against all PRs, even those with multiple distinct decryption keys. CTT is made with a simple combination of schemes that are absolutely necessary. It makes broadcasting or multicast services secure.

Bistatic multi-input multi-output (MIMO) radar has the capability of measuring the transmit angle from the receiving array, which means the existence of information redundancy and benefits data association. In this paper, a data association decision for bistatic MIMO radar is proposed and the performance advantages of bistatic MIMO radar in data association is analyzed and evaluated. First, the parameters obtained by receiving array are sent to the association center via coordinate conversion. Second, referencing the nearest neighbor association (NN) algorithm, an improved association decision is proposed with the transmit angle and target range as association statistics. This method can evade the adverse effects of the angle system errors to data association. Finally, data association probability in the presence of array directional error is derived and the correctness of derivation result is testified via Monte Carlo simulation experiments. Besides that performance comparison with the conventional phased array radar verifies the excellent performance of bistatic MIMO Radar in data association.

Although many 3D structures have been solved for proteins to date, functions of some proteins remain unknown. To predict protein functions, comparison of local structures of proteins with pre-defined model structures, whose functions have been elucidated, is widely performed. For the comparison, the root mean square deviation (RMSD) has been used as a conventional index. In this work, adaptive deviation was incorporated, along with Bregmann Divergence Regularized Machine, in order to detect analogous local structures with such model structures more effectively than the conventional index.

Speeded up robust features (SURF) can detect/describe scale- and rotation-invariant features at high speed by relying on integral images for image convolutions. However, the time taken for matching SURF descriptors is still long, and this has been an obstacle for use in real-time applications. In addition, the matching time further increases in proportion to the number of features and the dimensionality of the descriptor. Therefore, we propose a fast matching method that rearranges the elements of SURF descriptors based on their entropies, divides SURF descriptors into sub-descriptors, and sequentially and analytically matches them to each other. Our results show that the matching time could be reduced by about 75% at the expense of a small drop in accuracy.

There exist two intrinsic issues in multiple-shot person re-identification: (1) large differences in camera view, illumination, and non-rigid deformation of posture that make the intra-class variance even larger than the inter-class variance; (2) only a few training data that are available for learning tasks in a realistic re-identification scenario. In our previous work, we proposed a local distance comparison framework to deal with the first issue. In this paper, to deal with the second issue (i.e., to derive a reliable distance metric from limited training data), we propose an adaptive learning method to learn an adaptive distance metric, which integrates prior knowledge learned from a large existing auxiliary dataset and task-specific information extracted from a much smaller training dataset. Experimental results on several public benchmark datasets show that combined with the local distance comparison framework, our adaptive learning method is superior to conventional approaches.

In this paper, we propose a novel approach for multiple-shot people re-identification. Due to high variance in camera view, light illumination, non-rigid deformation of posture and so on, there exists a crucial inter-/intra- variance issue, i.e., the same people may look considerably different, whereas different people may look extremely similar. This issue leads to an intractable, multimodal distribution of people appearance in feature space. To deal with such multimodal properties of data, we solve the re-identification problem under a local distance comparison framework, which significantly alleviates the difficulty induced by varying appearance of each individual. Furthermore, we build an energy-based loss function to measure the similarity between appearance instances, by calculating the distance between corresponding subsets in feature space. This loss function not only favors small distances that indicate high similarity between appearances of the same people, but also penalizes small distances or undesirable overlaps between subsets, which reflect high similarity between appearances of different people. In this way, effective people re-identification can be achieved in a robust manner against the inter-/intra- variance issue. The performance of our approach has been evaluated by applying it to the public benchmark datasets ETHZ and CAVIAR4REID. Experimental results show significant improvements over previous reports.

A new scheme based on multi-order visual comparison is proposed for full-reference image quality assessment. Inspired by the observation that various image derivatives have great but different effects on visual perception, we perform respective comparison on different orders of image derivatives. To obtain an overall image quality score, we adaptively integrate the results of different comparisons via a perception-inspired strategy. Experimental results on public databases demonstrate that the proposed method is more competitive than some state-of-the-art methods, benchmarked against subjective assessment given by human beings.

Many subjective assessment methods for video quality are provided by ITU-T and ITU-R recommendations, but the differences among these methods have not been sufficiently studied. We compare five subjective assessment methods using four quantitative performance indices for both HD and QVGA resolution video. We compare the Double-Stimulus Continuous Quality-Scale (DSCQS), Double-Stimulus Impairment Scale (DSIS), Absolute Category Rating method (ACR), and ACR with Hidden Reference (ACR-HR) as common subjective assessment methods for HD and QVGA resolution videos. Furthermore, we added ACR with an 11-grade scale (ACR11) for the HD test and Subjective Assessment of Multimedia Video Quality (SAMVIQ) for the QVGA test for quality scale variations. The performance indices are correlation coefficients, rank correlation coefficients, statistical reliability, and assessment time. For statistical reliability, we propose a performance index for comparing different quality scale tests. The results of the performance comparison showed that the correlation coefficients and rank correlation coefficients of the mean opinion scores between pairs of methods were high for both HD and QVGA tests. As for statistical reliability provided by the proposed index, DSIS of HD and ACR of QVGA outperformed the other methods. Moreover, ACR, ACR-HR, and ACR11 were the most efficient subjective quality assessment methods from the viewpoint of assessment time.

This article presents an algorithm that solves an on-line version of the longest common subsequence (LCS) problem for two strings over a constant alphabet in O(d+n) time and O(m+d) space, where m is the length of the shorter string, the whole of which is given to the algorithm in advance, n is the length of the longer string, which is given as a data stream, and d is the number of dominant matches between the two strings. A new upper bound, O(p(m-q)), of d is also presented, where p is the length of the LCS of the two strings, and q is the length of the LCS of the shorter string and the m-length prefix of the longer string.

For measuring the similarity of biological sequences and structures such as DNA sequences, protein sequences, and tertiary structures, several compression-based methods have been developed. However, they are based on compression algorithms only for sequential data. For instance, protein structures can be represented by two-dimensional distance matrices. Therefore, it is expected that image compression is useful for measuring the similarity of protein structures because image compression algorithms compress data horizontally and vertically. This paper proposes series of methods for measuring the similarity of protein structures. In the methods, an original protein structure is transformed into a distance matrix, which is regarded as a two-dimensional image. Then, the similarity of two protein structures is measured by a kind of compression ratio of the concatenated image. We employed several image compression algorithms, JPEG, GIF, PNG, IFS, and SPC. Since SPC often gave better results among the other image compression methods, and it is simple and easy to be modified, we modified SPC and obtained MSPC. We applied the proposed methods to clustering of protein structures, and performed Receiver Operating Characteristic (ROC) analysis. The results of computational experiments suggest that MSPC has the best performance among existing compression-based methods. We also present some theoretical results on the time complexity and Kolmogorov complexity of image compression-based protein structure comparison.

This work first investigates two existing check node unit (CNU) architectures for LDPC decoding: self-message-excluded CNU (SME-CNU) and two-minimum CNU (TM-CNU) architectures, and analyzes their area and timing complexities based on various realization approaches. Compared to TM-CNU architecture, SME-CNU architecture is faster in speed but with much higher complexity for comparison operations. To overcome this problem, this work proposes a novel systematic optimization algorithm for comparison operations required by SME-CNU architectures. The algorithm can automatically synthesize an optimized fast comparison operation that guarantees a shortest comparison delay time and a minimized total number of 2-input comparators. High speed is achieved by adopting parallel divide-and-conquer comparison operations, while the required comparators are minimized by developing a novel set construction algorithm that maximizes shareable comparison operations. As a result, the proposed design significantly reduces the required number of comparison operations, compared to conventional SME-CNU architectures, under the condition that both designs have the same speed performance. Besides, our preliminary hardware simulations show that the proposed design has comparable hardware complexity to low-complexity TM-CNU architectures.

It is said that technology comes out from humanity. What is humanity? The very definition of humanity is emotion. Emotion is the basis for all human expression and the underlying theme behind everything that is done, said, thought or imagined. Making computers being able to perceive and respond to human emotion, the human-computer interaction will be more natural. Several classifiers are adopted for automatically assigning an emotion category, such as anger, happiness or sadness, to a speech utterance. These classifiers were designed independently and tested on various emotional speech corpora, making it difficult to compare and evaluate their performance. In this paper, we first compared several popular classification methods and evaluated their performance by applying them to a Mandarin speech corpus consisting of five basic emotions, including anger, happiness, boredom, sadness and neutral. The extracted feature streams contain MFCC, LPCC, and LPC. The experimental results show that the proposed WD-MKNN classifier achieves an accuracy of 81.4% for the 5-class emotion recognition and outperforms other classification techniques, including KNN, MKNN, DW-KNN, LDA, QDA, GMM, HMM, SVM, and BPNN. Then, to verify the advantage of the proposed method, we compared these classifiers by applying them to another Mandarin expressive speech corpus consisting of two emotions. The experimental results still show that the proposed WD-MKNN outperforms others.

This paper proposes probabilistic pruning techniques for a Bayesian video face recognition system. The system selects the most probable face model using model posterior distributions, which can be calculated using a Sequential Monte Carlo (SMC) method. A combination of two new pruning schemes at the resampling stage significantly boosts computational efficiency by comparison with the original online learning algorithm. Experimental results demonstrate that this approach achieves better performance in terms of both processing time and ID error rate than a contrasting approach with a temporal decay scheme.