Research On Computational Aesthetic Evaluation,Modeling And Rendering Of Graphics And Images

With the rapid development of computer science and digital media technology,an increasing number of visual artworks are created,stored and propagated in digital form,becoming an indispensable part of human being's spiritual wealth.However,traditional aesthetic evaluation of visual artworks heavily relies on professional artists and evalu-ation experts,the evaluated results have a high degree of subjective uncertainty due to their different personal knowledge and experience.Also,it is impossible to manually evaluate the massive digital artworks.To alleviate this situation,computational aesthet-ics appears,aiming at quantitatively evaluating digital artworks by computer models.These models have a wide application prospect in many fields such as advanced image retrieval,image aesthetic evaluation and image style identification.Additionally,some graphic patterns,such as logos and ethnic patterns,must follow aesthetic design rules in their forms and structures.If we can utilize computers to analyze their aesthetic features and generate artistic patterns efficiently,the relevant studies are significant in reveal-ing aesthetic rules of artistic patterns quantitatively,modern pattern design and digital protection of ethnic artistic patterns.However,such studies are challenging for they require interdisciplinary knowledge in computer science,cognitive science,psychology,aesthetics and arts.In this thesis,we first choose two types of digital artworks from graphic and image filed,logos and Chinese ink paintings,and research on corresponding aesthetic variables construction and aesthetic quantitative evaluation metrics.Next,we develop an approach to model and render mandala patterns in Thangka art.The main focuses and contributions of this thesis are listed below.1.We develop computational aesthetic evaluation models of logos.First we con-struct some feature variables of balance,contrast and harmony in logos,and convert these design principles into mathematical expressions.Then,we conduct human ratings of perceived balance,contrast,harmony and aesthetics on selected logos.Correspond-ing linear regression models of three design principles are then obtained by regressing human evaluated scores on calculated feature variables.Finally,we regress human e-valuated aesthetics scores on model evaluated balance,contrast and harmony scores to obtain an aesthetic model.Experimental results show that our regression models can predict human judgments on perceived balance,contrast,harmony and aesthetics with significant correlation over the range between 0.851 and 0.964.2.We present a framework for computational aesthetic evaluation of Chinese ink paintings.Our framework first extracts some features variables on the color,texture,and composition in the selected Chinese ink paintings,and develops their mathematical expressions.Next,we collect human ratings on these feature variables and perceived aesthetics.Then,we regress human evaluated scores on the scores calculated by fea-ture variables to obtain the support vector regression(SVR)models of these extracted attributes.Finally,we regress human evaluated aesthetics scores on model evaluated col-or,texture and composition scores to develop an aesthetics model.Experimental results show that our SVR models of perceived color,composition,texture and aesthetics can predict human evaluations on the same dimensions with significant correlation over the range between 0.822 and 0.904.3.We propose an approach for modeling and rendering mandala patterns in Thang-ka.First we construct parameterized models of three kinds of motifs used in the interior mandalas of Thangka from Tibetan Buddhism,namely the star,crescent and lotus flower patterns.Varieties of interior mandala patterns are then generated by arranging motif-s in a layer-by-layer fashion based on the aesthetic rules used in hand drawn mandala thangkas.Experimental results show that our approach can efficiently generate various colorful interior mandala thangka patterns.From the theoretical perspective,our studies extend computational aesthetics to logos and Chinese ink paintings.From application perspective,our mandala pattern ren-dering system can not only re-construct existing mandala thangka patterns,but also gen-erate new concentric patterns.Potential applications of our work may range from graphic design,advertising design,decoration design,commercial booth design to digital enter-tainments.