Visual Perception And Spatial Optimization Based Artistic Image Synthesis

Artistic image synthesis is an important research direction in the field of artificial intelligence and non-photorealistic rendering.Based on the understanding of artistic image analysis and synthesis,we use data driven method,constraint with some aesthetic rules artists used in creating artistic image,and generate a reasonable layout with appropriate optimization algorithm.We conduct two work about layout optimization.One is beautifying handwriting characters based on the optimization of strokes and structure with a data-driven method.The other is gestalt based picture collage generation considering both semantic and color feature.Handwriting characters beautification usually refers to optimize strokes and topological structure of online or offline handwritten Chinese character with calligraphy aesthetic knowledge and experience.Although print form makes it easy to type characters,inputting handwritten character of personal style is better to convey emotion when writing electronic card and email in instant communication software.In addition,the effect of handwriting is bad for lack of practicing calligraphy.In order to improve the effect of handwriting,we present a method of beautifying handwriting based on data-driven.First,we introduce the formalization description of the handwriting's topological structure and strokes.Then,to beautify the spatial arrangement and strokes of the input character,we capture the writer's stability factors for the topology and strokes of users' handwriting,and choose suitable topological structure and strokes from the users' handwritten samples using"roulette wheel" method.In addition,we analyze some classical types of stroke connection described in calligraphy theory,and determine the strokes which need to consider the visual effect.Finally,we use curve fitting to optimize the relations of strokes to generate the final result that meets the requirements of calligraphy aesthetic.Picture collage usually refers to generate an informative and pleased artwork by placing pictures that come from user's photograph and Internet in a suitable method with some aesthetic rules.Traditional picture collage generation are usually created by skilled artists,which need to collect a lot of pictures in advance and professional image editing skills,it's a tedious manual design processes.In order to simplify this processes,we propose a method of generating picture collage based on Gestalt.We first design a vector field to guide the direction of picture layout.Then,we partition the region into a patch set along the streamlines with a divide-and-conquer strategy.To make the patch set more attractive,we use a saliency map of the source image to decide the patch's size.We get a lot of patch sets with different lightness threshold,and choose the best patch set which minimize the energy function constructed by Gestalt principles.Finally,we map the picture set to the patch set considering the semantic and color.A lot of experimental results and compare with other related work show that the proposed algorithm effectiveness in beautifying handwriting characters and generating picture collage.