Non-photorealistic rendering (NPR), a research about non-photorealistic images is a major field of research in image processing. Painterly rendering is a method that creates artistic images based on photo images and important in NPR. Recently, painterly rendering methods using evolutionary algorithm are studied. Those studies have intended to optimize the process of creating artistic images by using evolutionary algorithm. Most of those studies have focused on generating and placing strokes as a painting operation. On the other hand, some researchers proposed painterly rendering methods using existing art images. They have created painting images which have unique colors and textures of existing art images, called “painting style”. We propose a new method to create artistic images based on photo images and existing art images by using Genetic Algorithm (GA). Our method operates putting the “patches” on a canvas image repeatedly as a painting operation. We generate the “patches” by copying a part of the existing art images and put them on the canvas image in mutation of GA. We exchange pixels in the same region of two canvas images in crossover of GA. In the process of optimization, our method brings the canvas image close to the photo images. Our method evolutionarily creates the painting images which have the painting styles of existing art images.

抄録全体を表示

Under today's complicated and unstable society, it is becoming of special importance to make a rational decision from comprehensive point of view. Accordingly, many interests have been recently paid on the studies on multi-objective optimization. In this paper, we assert that it should be noted to distinguish the terminology between multi-objective optimization (MOP) and multi-objective analysis (MOA) from a rigid sense. Doing so, we can perform various tasks regarding MOA in more efficient manner. To promote such idea, we have first proposed a new method for MOA termed elite-induced MOEA that complementally combines conventional method and recent multi-objective evolutionary algorithm (MOEA). The former is applied to derive some Pareto optimal solutions as the elites while the latter to induce gradually the others targeting at the elites and dispersing on the Pareto front. Through this simple combination, the proposed method can derive the precise Pareto front and manipulate its distribution well. This provides a useful procedure for supporting decision making according to the preference of decision maker especially as a post-optimal analysis. These prospects are verified through numerical experiments.

抄録全体を表示

This paper proposes a multi-start optimization framework for the continuous multi-objective optimization problem (CMOP). Numerical solutions to a CMOP should have a satisfactory level of precision and coverage in order to give a fine representation of the Pareto set or the Pareto front. One successful approach in terms of precision is the descent method, guaranteeing convergence to a Pareto solution if the problem does not have non-Pareto-optimal critical points. However, the coverage of solutions produced by naive multi-start strategies, e.g., choosing initial parameters evenly or at random, would be unsatisfactory. This is because the correspondence between initial parameters, i.e., initial solutions, and weight vectors when a scalarization of objective functions is incorporated, and resulting solutions is usually unknown in advance and may be biased. Our proposal, Adaptive Weighted Aggregation (AWA), is a scalarization-based multi-start framework that employs two novel strategies taking coverage into account: 1) subdivision: a systematic initialization scheme utilizing the correspondence between the set of weight vectors and the Pareto set/front; 2) relocation: an iterative search mechanism for a well-placed Pareto solution that is equidistant from neighboring solutions with an adaptation of weight vectors. Alternately repeating the subdivision and the relocation, AWA approximates the Pareto set/front progressively from the boundary to the interior. We demonstrate the effectiveness of AWA by comparing it to conventional multi-start descent methods on 2- to 6-objective benchmark problems. Several aspects of the computational efficiency and the scalability of AWA are also discussed.

抄録全体を表示

Interactive Evolutionary Computation (IEC) is a method to optimize parameters with subjective evaluation by human user, and it is applicable with the creative tasks like composition or coloration, which computer system was not able to deal with before. In IEC process, evolutionary computation, which is one of the multi-point search algorithms, is used to find an optimal parameter set which user likes. Evolutionary computation is designed for the environment in which the system can use long term of computational time. However in situations of interactive parameter optimization, since input from human user is needed, foundation of a good solution in very short computational time is required considering fatigue of users. In these situations, evolutionary computation which tries to find global optima with long computational time is not necessarily suitable. In this research, to find a good solution in interactive situations, we propose a new search algorithm with two stages of multi-point search and one-point search. From the results of mathematical benchmark tests, we found that our method is effective in the environment with limited number of evaluations, which contains large segment of conventional IEC applications. Additionally, we also show the effectiveness of our method with a real world application.

抄録全体を表示

This paper proposes a new method of Estimation Distribution Algorithm (EDA) named Bayesian Optimization Algorithm with Mixture Distribution (BOA-MD) that employs mixture of multiple Bayesian Networks to solve complex problems. In order to solve complex problems that are modeled by multiple Bayesian networks with hidden variables, the original BOA needs a large computation cost to model multiple probabilistic structures as a large, complex Bayesian network.The BOA-MD tries to build multiple models of Bayesian networks considering hidden variables with Expectation Maximization (EM) method to express all the structures of probabilistic distribution.The mixture of Bayesian networks is composed of a hidden variable C and some Bayesian Networks. Each composed Bayesian network can express each problem structure of multiple distributions. We perform numerical experiments by two test functions: Cross-Trap function and Triple-Trap function. These two test functions are to represent problems with multiple distributions. BOA-MD can solve these test problems with smaller number of fitness evaluations and larger modeling overheads than those by BOA for Cross-Trap5 function. This is because BOA-MD needs large computation time to construct Mixture of Bayesian Network. The BOA-MD can solve the problem faster than the original BOA when the overhands of each fitness evaluation becomes larger. At Triple-Trap function, BOA-MD can detect better solution than BOA.

抄録全体を表示