Since the relation of labels in Binary Relevance (BR) is ignored, it is easy to cause the multi-label classifier to output not exist or less emergent labels in training data. The Multi-Label classification algorithm based on Bayesian Model (MLBM) and Markov Multi-Label classification algorithm based on Bayesian Model (MMLBM) were proposed. Firstly, to analyze the shortcomings of BR algorithm, the simulation model was established; considering the value of label should be decided by the attribute confidence and label confidence, MLBM was proposed. Particularly, the attribute confidence was calculated by traditional classification and the label confidence was obtained directly from the training data. Secondly, when MLBM calculated label confidence, it had to consider all the classified labels, thus some of no-relation or weak-relation labels would affect performance of the classifier. To overcome the weakness of MLBM, MMLBM was proposed, which used Markov model to simplify the calculation of label confidence. The theoretical analyses and simulation experiment results demonstrate that, in comparison with BR algorithm, the average classification accuracy of MMLBM increased by 4.8% on emotions dataset, 9.8% on yeast dataset and 7.3% on flags dataset. The experimental results show that MMLBM can effectively improve the classification accuracy when the label cardinality is larger in the training data.

A new soft subspace clustering algorithm was proposed to address the optimization problem for the projected subspaces, which was generally not considered in most of the existing soft subspace clustering algorithms. Maximizing the deviation of feature weights was proposed as the sub-space optimization goal, and a quantitative formula was presented. Based on the above, a new optimization objective function was designed which aimed at minimizing the within-cluster compactness while optimizing the soft subspace associated with each cluster. A new expression for feature-weight computation was mathematically derived, with which the new clustering algorithm was defined based on the framework of the classical k-means. The experimental results show that the proposed method significantly reduces the probability of trapping in local optimum prematurely and improves the stability of clustering results. And it has good performance and clustering efficiency, which is suitable for high-dimensional data cluster analysis.

The traditional statistical thresholding methods which directly construct the optimal threshold criterions by the class-variance have certain versatility, but lack the specificity of practical application in some cases. In order to select the optimal threshold for blood cell image segmentation and extract white blood cells nuclei, a simple and fast method based on cloud model was proposed. The method firstly generated the cloud models corresponding to white blood cells nuclei and blood cell image background respectively, and defined a new thresholding criterion by utilizing the hyper-entropy of cloud models, then obtained the optimal grayscale threshold by the maximization of this criterion, finally achieved blood cell image thresholding and white blood cells nuclei extraction. The experimental results indicate that, compared with the traditional methods including maximizing inter-class variance method, maximizing entropy method, minimizing error method, minimizing intra-class variance sum method, and minimizing maximal intra-class variance method, the proposed method is suitable for blood cell image thresholding, and it is reasonable and effective.

In order to improve the convergence performance of Quantum-behaved Particle Swarm Optimization (QPSO) algorithm, this paper proposed an improved QPSO algorithm which was called RE-QPSO based on the random evaluation strategy. The new algorithm evaluated the innovation of particles by using a random factor and improved the ability of the particles to get rid of the local optima. Fixed value strategy and linear decreasing strategy were proposed for controling the theunique parameter of QPSO algorithm and they were tested on six benchmark functions. According to the test results, some conclusions concerning the selection of the parameter were drawn.

Centralized CBR (Case-Based Reasoning) is faced with problems such as massive data maintenance cost, high node load and central failure. To tackle these problems, COCO (Distributed Hash Table (DHT) based Collaborative Web Service Composition) was proposed. COCO mappped the workflow and QoS properties of a composite service into a one-dimensional key using Hash function and Space-Filling Curve (SFC). Keys were used for queries of the existing composite services that satisfied user request in a Peer-to-Peer (P2P) fashion in the underlying DHT Overlay. One successful query returned composite services satisfying both functional and non-functional requirements. The experimental results show that COCO has scalable performance on both query delay and query hit rate, indicating COCO's applicability in large-scale network computing environment.

To select the optimal threshold for image segmentation, a new method based on local complexity and local difference was proposed. Firstly, the local grayscale features of a given image were generated, including local complexity and local difference. Next, the new feature matrix was constructed using local feature fusion. Then, an automatic threshold was defined based on the mean and standard deviation of feature matrix, and the image transition region was extracted. Finally, the optimal grayscale threshold was obtained by calculating the grayscale mean of transition pixels, and the binary result was yielded. The experimental results show that, the proposed method performs well in transition region extraction and thresholding, and it is reasonable and effective. It can be an alternative to traditional methods.

The multi-side covering algorithm is designed guided by the idea of divide-and-conquer to the mass high-dimensional data. According to the sum of the absolute value of the component deviation, subsets of attributes were selected to construct respective covering domains for different parts of training samples, thus reducing the complexity of learning. But the selection of initial attribute set should be acquired by experience or experiments. In order to reduce the subjectivity with the selection of initial attribute set and the complexity with the regulation of attribute set, the relief feature selection approach was used to ensure the optimal feature subset that can be appropriate for different data sets, build a hierarchical overlay network, and experiment on the actual data set. The experimental results show that this algorithm is provided with higher precision and efficiency. Therefore, the algorithm can effectively achieve the classification of the complex issues.