Research On Photography Location Recommendation Based On Scene Semantic Segmentation

The goal of photography location recommendation is to suggest the best layout of people location in photographic scenes for the users.Photography location recom-mendation can help users to achieve human and scenery photos with higher aesthetic value.Such technology can be applied into many fields such as social networks,the automatic generation of high-quality content in image processing software and photog-raphy in travel records.Nowadays the photography location recommendation mainly relies on the hand-craft image features and the methods based on deep learning tech-nology.Many researchers have studied the feature fusion of 2D image aesthetics but lack of the depth features.Most of the photography scenes contain rich semantic in-formation.However the existing photography location recommendation technology lacks the utilization of scene semantic information.Our work is inspired by the the-ory of computational aesthetics and propose a new method on photography location recommendation based on scene semantic segmentation.This technology can extract semantic labels of the objects in photography scenes.Our work propose a new multi-scale aesthetic assessment model including three-dimensional information.Based on image semantic segmentation,the result of our aesthetic assessment can recommend users where to stand in the photography scenes and achieve high-quality photos.This paper studies the scene semantic segmentation and image aesthetic assess-ment technology.The main work includes scene semantic segmentation technology based on saliency estimation and photography location recommendation technology based on scene semantic segmentation.1.We improve a scene semantic segmentation method that integrates seed gener-ation and saliency estimation.Semantic segmentation can optimize the perfor-mance of the aesthetic assessment model.Human and scenery photos contain rich semantic information and highly influenced photography aesthetics.In this paper,we study the scene semantic segmentation technology based on the con-volutional neural network.Our work utilizes global average pooling based seed generation and semantic labelling based saliency estimation to achieve scene se-mantic information.The scene semantic information generated by this technique can enhance image aesthetic assessment.2.We propose a new image aesthetic assessment model based on scene seman-tic segmentation.This model can evaluate the aesthetic quality of the human and scenery photos and provide photography location recommendation for users.Most of the photography scenes contain three-dimensional information.The tra-ditional aesthetic assessment models only integrate two-dimensional image fea-tures,but lack three-dimensional information.Our work studies the technology of image aesthetics assessment and propose a new image aesthetic assessmen-t model based on scene semantic segmentation.We integrate the photography computational aesthetics,image aesthetics estimation and the complex saliency of the image.This technology can suggest the users where to stand in photog-raphy scenes and help solve the problems of personal layout optimizations in photography.Based on scene semantic segmentation and image aesthetic assessment model,we implement a photography location recommendation prototype system which applied in mobile platform.With the combination of high-speed computation in server system and good performance in client system,personal users can achieve photography loca-tion recommendation based on scene semantic segmentation.