Research On Automatic Recommendation In Story Visualization

In the Internet era, computer technology was popularized to thousands ofhouseholds. Science and technology are shown in front of people more and moreintuitively, changing our life and the way we communicate silently. Take storytellingas an example, the traditional storytelling plays an important way in childreneducation by teachers and parents; however, it only uses oral language or writtenlanguage to achieve its goal. Nowadays, story visualization provides the third way tostorytelling-pictures. During the process, pictures, as an auxiliary means ofcommunication, are got by using story visualization, improve the communicationefficiency. In this paper, based on the previous work, we improve the efficiency andeffectiveness of image generation using image feature extracted from huge amounts ofimages. The main contribution of this work can be summarized follows:Firstly, we design and implement a picture recommendation algorithm based onpicture’s semantic information. According to the key words, our system can obtainpictures from picture library, sort these results automatically, and then show theresults to user. This algorithm considers the key words’ semantic information,including the image’s main color, significant aggregation, color aggregation, and theoverall statistical law of the search results, and gives the final recommended results.Secondly, we design the common scene layout, and realize a scene layoutrecommendation algorithm. Users need to determine the scene layout and object’slocations in the scene manually before. By using data mining classification technique,our system calibrates the keyword’s layout automatically according to the specifickeywords. Users can use the recommended result directly or with a little trimming.At last, we design and implement a distributed system of mass image features.Our system provides a tool to download pictures from website using multi-thread, tohelp user grab millions of pictures conveniently in preprocessing stage. Since oursystem is based on distributed computing, it can complete the image feature extractionefficiently. In addition, the system also provides fast access to image features.