Android Platform Based Research On Large Scale Face Recognition System

The last several decades have witnessed significant advancements and extensiveground breaking results in face recognition technology. While it has been widelyemployed in a plethora of applications, some harsh conditions remain as openchallenges. As the advent of the mobile era, mobile smart phones play a more andmore important role on human lifes, so researchers pay more attention on facerecognition technologies under mobile environment. However, ther are somecharicteristics of face recognition under mobile environment. This thesis discussesthose charicteristics of face recognition under mobile phones, and proposes somesolutions to two key issues, illumination and large scale, in terms of system andperformance. Based on those work, some prototype systems are developed fordemonstration. The main work includes:(1) Research on illumination preprocessing under mobile environment. Since theillumination under mobile face recognition ranges from explosure tocompletely dark, and few content solutions to the illumination variation, akey issue on face recognition, have been put through, this thesis proposed anovel method named mobile Gamma intensity correction (M-GIC) based onsensor data collected from the light sensor on mobile phones. This methodcan effectively balance the algorithm performance and system performanceon mobile devices which have limited computation capacity and energyresource. This idea serves as a modest spur to further research on illuminationcpmpensation of mobile face recognition.(2) Construction of standard face database specific for research on mobile facerecognition (HNU-MFD). This database includes22kinds of differentillumination scenes and others and owns the following charicteristics:1)collected under actual scenes;2) collected by mobile devices and undermobile environments;3) affiliated with text data collected from sensors.Strict collecting criterions and sufficient baseline experiments guarantee theeffectiveness of the database. This database is open and serves as the basicdata for further research on mobile face recognition.(3) Research on large-scale face recognition. The thesis proposes a method forlarge-scale face recognition based on cluster algorithms to improve both the speed and recognition rate. This method can reduce the worklord ofcomputation based on distributed architecture and distributed database.K-means cluster is adopted in this thesis and experiments show the speed isaccelerated effectively without sacrificing the recognition rate. Sincemethods based on cluster algorithms are not related with the face fature, theyown the decent compatibility and expansibilty.(4) Design and implementation of two forms and three models of facerecognition system based on Android platform. This thesis analyzes thesignificance and application scenarios of mobile face recognition. Two forms,by taking picture and by supervision, are developed as well as three models,online picture, online supervison and offline supervison. Experomentalresults show the solutions proposed in this thesis are effective to satisfy theactual requirements and serve as basic for further research on facerecognition under mobile environment.