Narrow Results

This paper proposes an efficient approach for human face detection and exact facial features location in a head-and-shoulder image. This method searches for the eye pair candidate as a base line by using the characteristic of the high intensity contrast between the iris and the sclera. To discover other facial features, the algorithm uses geometric knowledge of the human face based on the obtained eye pair candidate. The human face is finally verified with these unclosed facial features. Due to the merits of applying the Prune-and-Search and simple filtering techniques, we have shown that the proposed method indeed achieves very promising performance of face detection and facial feature location.

Defocus blur detection aims at separating regions on focus from out-of-focus for image processing. With today’s popularity of mobile phones with portrait mode, accurate defocus blur detection has received more and more attention. There are many challenges that we currently confront, such as blur boundaries of defocus regions, interference of messy backgrounds and identification of large flat regions. To address these issues, in this paper, we propose a new deep neural network with both global and local pathways for defocus blur detection. In global pathway, we locate the objects on focus by semantical search. In local pathway, we refine the predicted blur regions via multi-scale supervisions. In addition, the refined results in local pathway are fused with searching results in global pathway by a simple concatenation operation. The structure of our new network is developed in a feasible way and its function appears to be quite effective and efficient, which is suitable for the deployment on mobile devices. It takes about 0.2s per image on a regular personal laptop. Experiments on both CUHK dataset and our newly proposed Defocus400 dataset show that our model outperforms existing state-of-the-art methods.

Deep learning technology has greatly improved the performance of target tracking, but most recently developed tracking algorithms are short-term tracking algorithms, which cannot meet the actual engineering needs. Based on the Siamese network structure, this paper proposes a long-term tracking framework with a persistent tracking capability. The global proposal module extends the search area globally through the construction of a feature pyramid. The local regression module is mainly responsible for the confidence evaluation of the candidate regions and for performing more accurate bounding box regression. To improve the discriminative ability of the regression network, the error samples are eliminated by synthesizing the temporal information and are then classified through a verification module in advance. Experiments on the VOT long-term tracking dataset and the UAV20L aerial dataset show that the proposed algorithm achieves state-of-the-art performance.

Chaotic Ant Swarm (CAS) is an optimization algorithm based on swarm intelligence theory, which has been applied to find the global optimum solution in search space. However, it often loses its effectiveness and advantages when applied to large and complex problems, e.g. those with high dimensions. To resolve the problems of high computational complexity and low solution accuracy existing in CAS, we propose a Disturbance Chaotic Ant Swarm (DCAS) algorithm to significantly improve the performance of the original algorithm. The aim of this paper is achieved by three strategies which include modifying the method of updating ant's best position, neighbor selection method and establishing a self-adaptive disturbance strategy. The global convergence of the DCAS algorithm is proved in this paper. Extensive computational simulations and comparisons are carried out to validate the performance of the DCAS on two sets of benchmark functions with up to 1000 dimensions. The results show clearly that DCAS substantially enhances the performance of the CAS paradigm in terms of computational complexity, global optimality, solution accuracy and algorithm reliability for complex high-dimensional optimization problems.

We study in this paper the partitioning of the constraints of a temporal planning problem by subgoals, their sequential evaluation before parallelizing the actions, and the resolution of inconsistent global constraints across subgoals. Using an ℓ₁-penalty formulation and the theory of extended saddle points, we propose a global-search strategy that looks for local minima in the original-variable space of the ℓ₁-penalty function and for local maxima in the penalty space. Our approach improves over a previous scheme that partitions constraints along the temporal horizon. The previous scheme leads to many global constraints that relate states in adjacent stages, which means that an incorrect assignment of states in an earlier stage of the horizon may violate a global constraint in a later stage of the horizon. To resolve the violated global constraint in this case, state changes will need to propagate sequentially through multiple stages, often leading to a search that gets stuck in an infeasible point for an extended period of time. In this paper, we propose to partition all the constraints by subgoals and to add new global constraints in order to ensure that state assignments of a subgoal are consistent with those in other subgoals. Such an approach allows the information on incorrect state assignments in one subgoal to propagate quickly to other subgoals. Using MIPS as the basic planner in a partitioned implementation, we demonstrate significant improvements in time and quality in solving some PDDL2.1 benchmark problems.

In this work a co-evolutionary approach is used in conjunction with Genetic Programming operators in order to find certain transition rules for two-step discrete dynamical systems. This issue is similar to the well-known artificial-ant problem. We seek the dynamic system to produce a trajectory leading from given initial values to a maximum of a given spatial functional.

This problem is recast into the framework of input-output relations for controllers, and the optimization is performed on program trees describing input filters and finite state machines incorporated by these controllers simultaneously. In the context of Genetic Programming there is always a set of test cases which has to be maintained for the evaluation of program trees. These test cases are subject to evolution here, too, so we employ a so-called host-parasitoid model in order to evolve optimizing dynamical systems.

Reinterpreting these systems as algorithms for finding the maximum of a functional under constraints, we have derived a paradigm for the automatic generation of adapted optimization algorithms via optimal control. We provide numerical examples generated by the GP-system MathEvEco. These examples refer to key properties of the resulting strategies and they include statistical evidence showing that for this problem of system identification the co-evolutionary approach is superior to standard Genetic Programming.