Topology-varying Pattern Generation And Exploration

In pattern design,it is usually necessary to adjust the number,position,size and direction of elements,as well as the distance and angle between elements,so as to arrange and combine elements into an entirety with harmonious and beautiful structure.Manually designing graphical patterns is a time-consuming task and requires designers having certain professional foundation.Therefore,it is very significant to investigate high-efficiency techniques to aid pattern generation and browsing.Most of the existing methods are based on the given samples,which are changed to get similar variations.However,existing methods can only generate new patterns between the number of elements and patterns with the same topology.Different from these methods,this paper generates new patterns from samples with different number of elements and topological structures,and conducts smooth browsing between the generated patterns.The difficulty is embodied in three aspects:(1)building a good corresponding relation between patterns with different topological structure,(2)defining a metric for new patterns such that it can sample design and the structure characteristics of different fusion,but also the introduction of new change,(3)finding new patterns via a complicated optimization.In order to solve the above problems,this paper first constructs a diagram model for the pattern,which describes the structural characteristics of the pattern,including the location,size and direction of elements,the relationship between elements,and the relationship between relations and other information.Then,a discrete optimization model is proposed to establish a good correspondence between patterns.Next,an energy equation is constructed by combining the characteristics of the two patterns.In order to obtain various new patterns,this paper uses the reversible jump Markov Chain Monte Carlo algorithm to sample the required results from the energy equation.Finally,in order to display the generated new patterns well,this paper takes them as points in a high-dimensional space and projects them onto a two-dimensional manifold.By controlling the manifold,interpolation and smooth browsing between the generated patterns can be achieved.A large number of experiments show that the proposed method can effectively generate rich new patterns among samples with different topologies.At the same time,the corresponding proposed browsing method can effectively interpolate all the patterns smoothly and facilitate users to select novel patterns.