Research And Implementation Of Image Style Processing For Scratch Platform

With the emergence of visual programming tools,the content of information technology education in primary and secondary schools has changed.People pay more and more attention to the training of children's programming ability.At present,Scratch is the preferred language for teenagers to learn programming.Students need to choose the appropriate background image for the stage in scratch programming.The background image can not only provide the appropriate environment for the role's activities but also directly express the author's emotion and personality.However,there is no function related to image style on the current mainstream Scratch platform.The image style in all platforms is single,and an image has only one style.Considering that the background images with rich and diverse styles are conducive to express different emotions and enhance the visual effect of the works,so as to improve students'interest in learning on Scratch.Therefore,this thesis studies the image style processing of the Scratch platform and provides students with two kinds of Scratch background style processing methods:filter and art style transformation.To solve the problem that the content of the transformed image does not match the chosen style when the art style transformation tool is directly introduced into the scratch platform,we design a Scratch image art style conversion function based on image classification.Through the scene recognition of Scratch background,it helps students to match the art style suitable for this scene.In the research of Scratch image classification,we propose the network model based on the fusion of residual unit and dual attention mechanism,and design Scratch dataset construction module based on data augmentation method and self-adjusting learning rate module.Finally,we conduct comparative experiments to prove the effectiveness of the three improvements.To solve the problem of the blur of image details and degradation of image quality caused by directly using the existing art style transformation methods for the low-quality Scratch image,we design a Scratch image art style transformation scheme based on super-resolution reconstruction preprocessing.This scheme mainly includes Scratch image super-resolution reconstruction preprocessing research and art style model library construction based on the model iteration method.In the research of super-resolution reconstruction,we optimize the super-resolution reconstruction method based on the generative adversarial networks,which is mainly to remove the batch normalization layer in the generation network and embed the designed channel attention mechanism in the residual layer.Finally,we conduct comparative experiments to prove the effectiveness of the two research works.Finally,from the perspective of software engineering,we design and implement an image style processing system based on B/S architecture for the Scratch platform.In addition,we verify that the system achieves the expected effect through the functional test and performance test of the system,and the system can help students better complete the Scratch works.