Study Of Perception-motivated ASCⅡ Art Synthesis Method

With the development and popularity of text-based communication, using ASCII artcharacters to represent a variety of images is becoming more and more meaningful. Aneffective ASCII are generation algorithm must be able to handle such kinds of affects, likeslight deformation, noise and line misalignment et al. For this purpose, researches proposedmany ASCII are generation algorithm, but for generating ASCII art image from arbitraryimages is still a challenging task. To this end, this paper proposes a fast structure-based ASCIIart generation model for any image. The main work is as follows:1. A1D phase congruency model is introduced in this thesis, then lead to a2D phasecongruency model. After that, this thesis make a deep analysis to this2D model, then proposea multi-orientation phase congruency specify for our ASCII art application. Thismulti-orientation model can not only faithfully extract the main structure which is sensitive interms of visual system of human perception from arbitrary images, but also effectivelysuppress insensitive noisy texture base on human perception.2. In this thesis, a deeply survey is done for the existing methods for image similaritymetrics and analyze the feasibility of these methods as one step of ASCII art generation model.It can be found through the experiments that none of them can directly apply to ASCII artgenerating application. Thus, this paper propose a novel point-to-area(PTA) method as thematching strategy, then use Earth Mover’s Distance(EMD) to calculate the distance betweentwo feature vectors. Combining with the multi-orientation phase consistency model, theyconstitute our image similarity metrics in ASCII art generation.3. A dynamic programming based optimization is proposed to simultaneously determinethe optimal characters to match and their optimal placement locations. It decompose the main2D optimization problem into several1D optimization sub-problems. This way can greatlyreduce the time consuming on calculation, while the optimum sequence of characters can beobtained for each sub-problem.4. The comparison with HOG,FSIM and AISS these three methods and the user studyprove that the results in this paper in terms of visual aesthetic and timing performance are outperformance.