Content-aware Photo Collage Using Twinning-circle Packing

High quality photos have emerged in large numbers under the evolution of hand-held and wearable devices in recent years. Meanwhile, the demands of personalized de-sign have been increased significantly following with the improvement of the aesthetic level, which could be the result of human advancement. However, affected by the in-formation explosion, it is complicated to obtain information that people truly interested in from big data, especially presenting information according to personalizing forms. Facing the issue above, various kinds of image summarization technologies have been developed. At the same time, image summarization technology, which comes from picture collage, could be the directly and effective method to display large quantities of images comparing with the traditional technologies.Picture collage technology is defined as a solution by displaying the significant information from original images as much as possible in the limited bounded display regions in order to make an aesthetic collage from given pictures. To achieve this purpose, most of the traditional picture collage technologies are operated through op-timized methods, which particularly mean adjusting the orientation and position of showing pictures areas. Because the objective function is usually a high-dimensional and non-convex, many local extremes could lead to the extremely complicated opti-mized processes. Indeed, traditional picture collage technologies are highly consumed computations and might emerge unsatisfied results.This paper presents picture collage technique based on a twinning-circle packing algorithm that could avoid the drawbacks of optimized method collage. Firstly, salien-cy detection is adopted to obtain the important information of pictures, which includes the relative size and the shape of the display sub region. Here we transform the collage process to a packing process so as to present the twinning-circle packing algorithm that based on centroidal Voronoi tessellation. This process could avoid the complex opti-mization steps and make the whole process more intuitive. What’s more, the general-ized Voronoi diagram could solve the packing problem effectively. More importantly, the twinning-circle algorithm could bring out the concept of twinning-circle that drives the sub region of the elliptical shape to display important information of similar shape. The important information could be displayed because of the sufficient use of limited region.For the purpose of maintaining more important information, this paper presents a boundary optimization algorithm based on Seam Carving as following:an optimal boundary is adopted to replace the original straight one that makes the transition more natural. At the same time, the important information could be retained integrally. Blue-noise point set generation based initialization and a series of heuristic rules are adapted to obtain a more efficient initialization, which could lead to a fast convergence. It is achievable to receive a stunning collage result with competitive time consuming.Twinning circle packing could be imposed in the picture collage problems as well as the conventional application areas problems, especially for packing problems, ma-terial stuffing issues and signal coverage problems.