Research On Intelligent Coding And Fast Retrieval Method Of Face Video Image

The traditional video surveillance system has the following two shortcomings: First,the amount of data is very large.The processing,analysis and viewing of massive data consumes a lot of manpower,material and financial resources.Second,the storage and transmission of video data is difficult.Massive video data often needs to be stored and passed to the background for analysis.The huge data makes data storage and delivery encounter bottlenecks and consumes a lot of hardware resources.Based on the above problems,this paper proposes a new intelligent video surveillance architecture to improve the intelligence level of video surveillance and reduce costs through information.The intelligent video surveillance architecture has the following salient features: 1.Faces are automatically detected in the natural scene in the video surveillance,without the active object being actively cooperated.2.The target face is intelligently encoded,effectively representing the face while saving storage and transferring resources.3.Fast retrieval is realized and the target face is efficiently retrieved in face coding database.This paper focuses on the research of intelligent coding and fast retrieval.The research content and innovations mainly include the following four parts:(1)An intelligent video surveillance architecture including front-end,back-end,and cloud is proposed.The video surveillance architecture is described in detail from the perspectives of system composition,workflow,and application views.The monitoring architecture includes video receiving,face detection,face grabbing,image enhancement,intelligent coding,fast retrieval and other modules,and each part cooperates to form a monitoring architecture.(2)This paper proposes the idea of applying binary gradient mode(BGP)to face intelligent coding,and proposes an improved and innovative method for the lack of intelligent level of BGP algorithm and the lack of deep extraction of image information.Specifically,three improved algorithms,such as BGP based heuristic information,cascaded BGP based on feature fusion,and cascaded BGP based on heuristic information,are proposed to apply the prior knowledge and heuristic information to face coding,and deep face information is obtained through multiple BGP extractions.The improved method described above significantly improves the effectiveness and accuracy of the encoding.(3)Two methods of fast retrieval based on pyramid coding and fast retrieval based on K-means clustering are proposed.The core of pyramid coding is to construct codes with different layer lengths for coarse-to-fine retrieval.K-means clustering method combines BGP's accurate representation of face with K-means clustering to improve retrieval efficiency under the premise of ensuring retrieval accuracy.