Research And Application Of Image Visual Attributes Transfer

Recently,the visual attributes transfer of images as a hot topic in the field of computer vision and image processing,the research institutions and scholars at home and abroad have carried out research experiments on it and different visual attributes are transferred,good transfer results have been achieved..The visual attributes transfer of images means to transfer the visual information of an image(such as color,texture,style,etc.)to another image,which are called reference and the original images respectively.In this thesis,the algorithm and application in three aspects include color transfer,style transfer and texture transfer are studied.The advantages and disadvantages of different algorithms and its application are analyzed.The related algorithm is extended to the images visual attribute transfer of some art schools and ethnic groups of Yunnan.This thesis presents the main algorithms of image visual attributes transfer,advantages and disadvantages of these algorithms and the scope of application of each algorithm have been summarized by experiments.Through advantages and disadvantages of these algorithms and the comparison of the experimental results,the design ideas and improved architecture of the images visual attribute transfer algorithm based on deep learning are analyzed and summarized.In view of the visual attribute transfer tasks of some art schools with Yunnan national characteristics,the appropriate visual attribute transfer is constructed after the analysis of its style characteristics.The main work of this thesis is:(a)Some image visual attribute transfer algorithms(including the classic visual attribute transfer algorithm and the visual attribute transfer algorithm based on deep learning)are implemented.Through analyzing the algorithm and experimental results,the advantages and disadvantages of these algorithms and their application areas are summarized.(b)The research and application of image visual attribute transfer is first extended to the school of Yunnan with ethnic characteristics,such as Yunnan heavy color painting and print woodcut engraving.According to the style characteristics of these schools,the best visual attribute transfer algorithm is constructed,and the good result has been obtained.(c)This thesis analyzes the effect of different visual attribute transfer algorithms on the style transfer of various art schools.By adjusting the parameters of the algorithms and comparing the experimental results after adjusting the parameters,the optimal parameter configuration is selected.(d)The image color histogram is applied to the objective evaluation of the color transfer,and the color transfer effect of different algorithms is evaluated by the distance of the color histogram and the correlation criterion.In this thesis,we compare several image visual attribute transfer algorithms and evaluate the visual attribute transfer results using objective evaluation criteria.An algorithm for visual attribute transfer of a class of images is obtained,and the parameters of the algorithm are adjusted to get the optimal parameter configuration.This work has taken a useful attempt in objective evaluation of images visual attribute transfer effect,avoids the deviations of subjective evaluation.