Research And Implementation Of Content-based Photo Management Platform

With the popularization and popularization of mobile phones in mass consumption,the camera function of smart phones has become a basic way for users to record their lives.Each person has hundreds or thousands of pictures in their mobile phones.Research and development of content-based intelligent picture management platforms enable efficient classification and management of pictures,which has certain theoretical research value and application value for the majority of smart phone users.The main work of this paper is to design a network structure that can be transplanted to the mobile terminal based on Hypotheses-CNN-Pooling(HCP)network.Based on this algorithm,a content-based photo management platform is implemented.There are two main ideas for improving the network.The first one is to use shared feature maps to reduce the network's computational complexity.The second is based on the first step HCP improvement method,in order to further reduce its occupation of resources,designed a MobileNet extraction HCP method of sharing characteristics,so that it can really be ported to the mobile terminal.The specific work is as follows:(1)In view of the complex structure of the HCP neural network and the fact that it takes up a lot of memory and cannot be ported to the mobile terminal,an HCP method for sharing feature maps is designed.HCP neural network is a multi-label neural network,but because of its high memory usage,it cannot be transplanted to mobile phones.In order to reduce its computational complexity,this paper applies the idea of Fast R-CNN shared feature maps to HCP neural networks.Sharing feature maps reduces the amount of calculations and reduces HCP computational complexity if the accuracy rate is substantially constant.The specific improvement method is to divide the shared neural network in the HCP into two parts.The former part is a feature network,and its role is to extract shared features.The latter part is a classifier.Its role is to achieve the feature map classification of the corresponding location of the candidate area.(2)In order to further reduce its resource consumption and memory consumption,this paper designs an HCP method to extract shared features using MobileNet.The HCP neural network uses VGG-16 and AlexNet as the shared network,but due to the large size of the VGG-16 network,even if the shared feature map is used,it cannot perform well on the mobile terminal,and the AlexNet accuracy rate is relatively low and does not meet the demand.In re,,sponse to this situation,this article replaces the shared neutal network with MobileNet based on the first step of HCP improvement.Because the MobileNet neural network has a small amount of computation and high accuracy,using MobileNet can reduce the computational complexity and memory footprint of the HCP network while keeping the accuracy rate constant,so that it can be applied to the mobile terminal.(3)Based on the improved HCP network,a content-based picture management platform is implemented.Users do not need to upload their photos to the server for image management.The platform is mainly based on picture content management pictures.The picture content mainly includes time information,location information,and picture scene content information in the picture Exif information.In addition,in order to facilitate users to use this management platform also realizes the function of displaying pictures according to the user's local gallery,while providing many small functions such as taking pictures,generating albums,and the like.The platform can intelligently manage the pictures in the user's photo album so that users can find useful information in the pictures.