In the existing person re-identification approaches based on sparse representation, l₁ regularization is generally utilized to approximate l₀-norm sparsity. Under the restricted isometry property (RIP) conditions, the l₀ regularization is equivalent to the l₁ regularization. However, it is difficult to meet the RIP conditions in person re-identification with many disturbing factors, such as cluttered background and object occlusion. In this paper, a group sparse representation method with mixed l₂/l_1/2-norm is proposed. The identical person image sequence in the gallery is regarded as a group, the intra-group structure is constrained by l₂-norm, and the inter-group structure is constrained by l_1/2-norm. The resulting model is more robust to the occlusion and cluttered backgrounds. The human body structure constraint is introduced to further enhance the discriminability of the proposed model. The person image is divided into several neighboring block regions. An adaptive sparse model of mixed l₂/l_1/2 norm is constructed for each region. Finally, the several sparse models are merged to identify persons. Experiments on PRID 2011 and iLIDS-VID datasets verify the effectiveness of the proposed model.

The recommendation algorithm based on matrix factorization has problems of data sparsity, cold start, poor anti-attack ability, etc. Therefore, a trust-based matrix factorization recommendation algorithm is proposed. Firstly, based on the principle of trust generation in social psychology, a reputation-based trust computation method is proposed to alleviate the trust data sparsity problem. Then, grounded on the principle of social homogenization, the user latent factor vector in the process of matrix factorization is extended by using the trust users to solve the rating data sparsity and new-user cold start problem. Meanwhile, social trust relationships are utilized to normalize the target function to improve the accuracy of the rating prediction. Experimental results on Epinions dataset show that the proposed method improves the recommendation precision greatly compared with the state-of-the-art methods, and it effectively solves the problems of data sparsity and cold start.

The attribute reduction of covering decision information systems is one of the most important problems of the rough set theory. In this paper, the decisions of covering decision information systems are characterized by coverings, and attribute reductions of the covering decision information systems are explored. The belief and plausibility functions from the evidence theory are employed to characterize attribute reductions in the covering decision information system. By plausibility function values of decision classes, the definitions of significance and relative significance of coverings are also developed. Then, an attribute reduction algorithm based on the evidence theory is proposed in the covering decision information system, and an example is adopted to illustrate the validity of the propose algorithm.

Entire-granulation rough sets can express explicit and implicit knowledge, as well as complexity, diversity and uncertainty of human cognition. Combined with classic rough set theory, several uncertainty indexes in entire-granulation rough sets are defined, including membership degree of entire-granulation, roughness degree of entire-granulation, dependence degree of entire-granulation for a single concept and dependence degree of entire-granulation for a decision system. The properties of these indexes are investigated, and the relations between these indexes and absolute attribute reducts of entire-granulation, attribute reducts of entire-granulation for a single concept and entire-granulation Pawlak reducts are indicated. The result is a theoretical foundation for attribute reduction and practical application of entire- granulation rough sets.

Basic operations of q-rung fuzzy sets are defined, and some of their properties are studied. By combining q-rung fuzzy sets and hesitant fuzzy sets, the concept of q-rung hesitant fuzzy sets is proposed. Then, some basic operations of q-rung hesitant fuzzy sets are defined and their properties are studied. Score functions and accuracy functions of q-rung hesitant fuzzy sets are defined. Based on the operations of q-rung hesitant fuzzy sets, the multi-criteria decision making model is constructed, and the feasibility and validity of the proposed model are verified by examples.

The overlapped region samples are divided through merging and dividing operation of sets based on the optimal partition concept, and thus the partition achieves minimal error. In this paper, the optimal partition of cover is introduced into the community detection, and a community detection algorithm based on optimal partition of covers(CDA_OPC) is proposed. The problem of community detection is converted to the optimal partition problem for the fixed coverage. In CDA_OPC, the covers can be constructed by utilizing the overlapped relationship between the nodes. Secondly, the optimal approximation of coverage is obtained through the merging and segmentation of covering subsets based on the concept of the optimal partition. Finally, the similarity between communities is calculated, and the multi-granularity community structure is finally formed through multi-level integration. Experimental results on real networks show that CDA _OPC is effective at community detection of networks.

The prediction of short-term bus passenger flow is the basis of the dynamic scheduling of the intelligent bus system. Therefore, according to the data characteristics of short-term passenger flow of bus, a short-term bus passenger flow prediction method based on string theory is proposed. A string invariants passenger flow prediction model(SI-PFPM) is constructed by simulating the string structure and the genetic algorithm is adopted to optimize the parameters of SI-PFPM. An AP clustering algorithm (DTW-AP) based on dynamic time warping(DTW) distance is proposed to perform clustering analysis on the passenger flow time series of short-time buses. SI-PFPM is employed to predict the short-term passenger flow data, and the predicted residual error is analyzed. The result shows that SI-PFPM is effective for the prediction of short-term bus passenger flow. Finally, the prediction performance of SI-PFPM is compared with the existing methods, and the effectiveness of SI-PFPM in short-term bus passenger flow prediction is verified.

Different hues, contrasts and blurriness of underwater images are caused by underwater environment, light attenuation and photography methods. The dark channel prior(DCP) or the maximum intensity prior(MIP) is often utilized in the underwater image restoration methods based on the image formation model(IFM). Low-quality restored images are produced by these methods, and they are easily disturbed by the complicated underwater environment. The method of image restoration using background light fusion and underwater DCP and image color balancing for underwater image enhancement is proposed in this paper. Firstly, a correct background light(BL) is estimated through the fusion of multiple candidate background lights.Then, an underwater dark channel prior(UDCP) is determined based on the statistics of a large number of high-quality(HQ) underwater images and accurate RGB transmission maps are finally obtained. The restored image in the RGB color model is transformed to a CIE-Lab color model, and the ‘L’ luminance component and color components ‘a’ ‘b’ are conducted with the normalized stretching and optimal modification respectively to further improve the brightness and contrast of the restored image. Various qualitative and quantitative assessments are applied to demonstrate that the proposed method is better than the state-of-the-art restoration methods in contrast, brightness and color.