Research On Photo Classification And Re-finding Technology For Mobile Devices

In recent years,with the continuous development of science and technology,mobile devices are widely used.Various major mobile phone manufacturers have successively launched mobile phones with high configuration,and the technology of mobile phones' taking photos is becoming more and more powerful.In view of the advantages of taking photos by mobile phones,such as convenient carrying,simple operation,timely sharing and high cost performance,people increasingly prefer to take photos with mobile phones.With the accumulation of time,the number of photos stored in the phone album is increasing day by day.How to effectively manage photos and improve the efficiency of re-finding photos is a worthy subject to study.We study this topic systematically,and analyze the insufficiency of traditional photo albums,the limitations of traditional photo classification and the shortcomings of traditional re-finding technology.Then we propose an album framework and the concrete implementation methods to classify photos based on specific events.What's more,we develop the album system.The main contributions of this paper are as follows:(1)This paper proposes an album framework classifying photos based on specific events.In this paper,the conceptual model of specific event is presented firstly,and then an album framework classifying photos based on specific events is proposed.The design idea of the framework is to classify the photos into clusters based on specific events,and then select a representative photo from each specific event to realize the visualization of each photo cluster.This framework reduces the time of scrolling up and down the screen to re-find a photo for users,and makes it easier for users to quickly locate the specific event to which the photo belongs.(2)This paper proposes a lightweight photo classification method based on specific events.After training,we find that photos belonging to the same specific event and different specific events have certain rules in terms of shooting time and shooting location,and then put forward the concept and calculation of spatio-temporal distance.According to the spatio-temporal distance of photos,a new method of classifying photos with low cost,low time complexity and in line with people's memory habits is developed.Since there is no available public photo set,we collect real personal photo sets for experiments.Experimental results show that the average accuracy of this classification method can reach 94.75% and the classification speed is fast.(3)This paper proposes a method to select a representative photo from a specific event to achieve the visualization of the photo cluster.We investigate and analyze the main features of the suitable photos which can visualize the photo clusters,and then grade photos by combining the image quality and memory factors.Finally,the photo with the highest scores is selected as the representative photo of a specific event to visualize the photo cluster.In this paper,the proposed method is compared with other selecting representative photos methods,and it is verified that the representative photos selected by our method are more consistent with people's memory habits and more representative.(4)This paper realizes the development of album system classifying photos based on specific events.We compare this album to the traditional album,the album classifying photos based on location information and the album classifying photos based on things in terms of the rate of re-finding photos.Experimental results show that the average re-finding rate of this album system is 0.856 times higher than traditional album,is 1.186 times higher than the album classifying photos based on location information,and is 0.896 times higher than the album classifying photos based on things.