Font Size: a A A

Design And Implementation Of Text Error Correction System Based On Text Extraction From Distributed Video Stream

Posted on:2021-07-04Degree:MasterType:Thesis
Country:ChinaCandidate:F Y LvFull Text:PDF
GTID:2518306308963819Subject:Computer technology
Abstract/Summary:PDF Full Text Request
In the age of multi-media,video has become an important information carrier,and the collection of text in the video is of great significance for analyzing video content and video retrieval and recommendations.Until now there are many ways of collecting video text information,such as optical character recognition and speech recognition.However,there is no such a method which can extract completely accurate text information.The OCR model is used to extract video text information in the project which this subject rely on.As the background in the video is complex,it is difficult to locate the text area and recognize text character with OCR model,and more errors occur in the collection result.Based on the requirements of video text extraction,this paper uses OCR model to extract video text information.For improving the accuracy of video text extraction,this paper design and train text error correction model,and finally design and implement a text error correction system based on text extraction from distributed video stream.Here is the main work in the paper:1.Analyze the system's functional and non-functional requirements carefully,use the successful MVC system designing patterns to design the system architecture.According to the analysis of system requirements,design and implement the functional modules of the system,provide users with the functions of video text extraction and text error correction,at the same time,design and implement the interaction logic between system and users.2.Design the way of storing and processing system data.The distributed file storage system HDFS is used to store user's video file,and the distributed message queue Kafka is used to decouple the flow of system data processing.Design and implement the user's video information table to save video information and reduce the redundancy of video files.Design and implement the video's OCR-Text table to store the output text of OCR model.Design and implement the video's text-correction-result table to store the corrected video text.3.Design and train a text error correction model based on the seq2seq framework.According to the characteristics of text error correction task,copy mechanism is added to the Transformer model.At the same time,text error detection task is added as a multi-task learning task.In the meantime,analyze the distribution of text information about users and crawl the corresponding corpus to train the text error correction model.The system has passed careful functional and non-functional tests,which proves the usability and effectiveness of the system,and the system is deployed and running in the real scenario.This paper has fully introduced the design and implementation of the system,and finally summarizes the work of this paper.Furthermore,this paper also describes the shortcomings of the system and introduces the future work.
Keywords/Search Tags:text correction, copy mechanism, multi-task learning, distributed message queue
PDF Full Text Request
Related items