Research Of Video Image Characters Recognition Algorithm And Its Implementation On FPGA

Video is a kind of datatype with different modalities;its main feature is a wealth of information,large of data volume and the high complexity of the algorithm.The development of the hardware quickly decreased the cost of video acquisition.At the same time,with the rapid development of the Internet,people get more video data than ever.The traditional method of hand-classified can't meet the retrieval requirement of the large sum of videos.The research and development of computer program and algorithm make it possible to finish content-based video retrieval automatically.The extraction and recognition of text in video is an important part of video retrieval.Video text retrieval involves the theory of digital image processing,pattern recognition,neural network and many other theories and methods.The thesis introduced and discussed several steps of character recognition in video,including text locating,the text extraction under complex background,text segment and character recognition.After simulated in MATLAB with different algorithm or method such as frame difference method,edge detectable algorithm,Morphological image processing method,Corner detection algorithm,Pulse coupled neural network algorithm and artificial neural network and finally decided to use frame difference method to finish the crude extract of character frame,then use corner detection algorithm and erosion and dilation operation to complete Text lines location roughly.The thesis puts forward an algorithm based on pulse coupled neural network and connected domain analysis method to improve the text extraction under complex background.As for character and word segmentation,the thesis uses the vertical projection method and made a discussion to conglutination problems.In order to get a better network for character recognition,this thesis simulates basic BP algorithm,vary learning rate BP algorithm and adding momentum factor BP algorithm.By analyzing the training time and convergence,this thesis combines vary learning rate BP algorithm with adding momentum factor BP algorithm,and use the coarse grids coupled with projection as feature vector.In this thesis,the author use an Altera FPGA as a hardware test platform to complete the hardware implementation of the system.First,in Quartus II development environment,using verilog to design video capture module,VGA module,SDRAM control module and median filtering module,and then using SOPC Builder to build embedded soft core system.Finally,in Nios II development environment,using C language to finish text locating,text extraction,character extraction and character recognition algorithm.The experimental results show that the characters recognition algorithm has a better behavior than before and the hardware implementation on FPGA basically meets the expectation,which provides experimental basis for the further research and development.