Research On Braille Image Recognition Based On Deep Learning

Purpose-With the expansion of China’s population and the increase in the number of blind people,the normal communication between blind people and Sighted people becomes more and more important,especially between blind students,teachers,parents and guardians of blind schools.In addition,a large number of precious early Braille works,books,test papers as time goes by,there will be a large number of damage and adhesion.Therefore,the Braille image recognition method,algorithm and system designed in this paper can help normal people to recognize Braille efficiently and correctly,so as to better understand and help the blind in life or learning,and to broaden the scope of employment of the blind,improve the quality of life of the blind and reproduce a large number of precious Braille materials have important significance.Method-In this paper,two Braille image recognition methods and models are proposed for three different Braille data sets.Among them,one method is based on twostage Braille image classification recognition.The specific recognition steps are as follows: firstly,the Braille image is preprocessed,then the Braille points are detected using Haar+Adaboost and other algorithms,and then the blind squares are divided according to the coordinate positions of the Braille points.Finally,the feature extraction and classification of the blind squares are carried out using the fusion model R-vit based on convolutional neural network and Transformer.The other method is based on Braille image recognition for target detection.Firstly,conduct Braille image pre-processing,then input all Braille point labels and position information into the target detection model Pyramid Net at once to get the final recognition and classification results.Findings-Two innovative Braille image recognition methods are put forward in this paper,and the accuracy rates of R-vit model and Pyramid Net model are both above98% on character Braille data set,one-sided Braille data set and double-sided Braille data set,which can meet the needs of the blind people in their daily life.At the same time,this paper uses Django framework to build a multi-functional Braille image recognition system SCOBR,which supports the recognition of character Braille data sets,single-sided Braille data sets and double-sided Braille data sets,and can realize the conversion from Braille to Chinese.Limitations of research-The Braille image recognition method and model proposed in this paper require relatively specific Braille data sets,that is,the Braille pictures have clear lines,specific Braille point coordinate position information should be calculated,and each blind square category should be marked.Practical implications-The work content of this paper has promoted the basic research in the field of information accessibility,enriched the research on Braille image recognition methods,and Transformer model has been successfully used in Braille image recognition research,laying a foundation for subsequent research.The SCOBR system not only facilitates the large number of people who need to translate Braille images into concrete results,but also satisfies the need for symmetric communication between blind people and Sighted people.Value-In this paper,we first create two data sets: character Braille data set and single-sided Braille data set.Compared with traditional Braille image recognition methods,this paper proposes two innovative Braille image recognition methods and models,one is based on two-stage Braille image classification recognition method,and proposes a fusion model R-vit based on convolutional neural network and Transformer.The other is based on Braille image recognition for target detection.Pyramid Net is proposed as the algorithm for target detection.