Research And Application Of Natural Scene Text Detection And Recognition Based On Neural Network

With the prosperity of computer vision technology,people are increasingly using smart hardware such as computers and mobile phones to detect and recognize text in nature,but detecting and recognizing texts in natural scenes is not an easy task.Therefore,how to effectively filter the information that is important to you in natural scenes becomes relevant and important.In recent years,a variety of text detection and recognition systems have been applied to various industries,and people's demand for text detection and recognition systems under natural scenes has also intensified.For example,the parking system of the property needs to accurately identify the license plate number to obtain the place of ownership and parking time;the blind navigation system needs to accurately identify the road sign,the roadside billboard,etc.to facilitate the travel of the blind and the driverless system of the car,etc.Therefore,text detection and recognition technology in natural scenes has great research significance and application value.The main research content of this paper is the detection and recognition of text in natural scenes.The research significance and background of natural scene text detection and recognition are elaborated.The status of natural scene text detection and recognition in recent years is introduced in detail.This paper studies the shortcomings of current mainstream text detection and recognition algorithms,proposes its own solution,and finally implements a system for text detection and recognition in natural scenes.The research content of the paper is as follows:(1)In terms of text detection,aiming at the problem that larger and longer texts are incomplete or misdetected in EAST text detection algorithms,So an improved EAST algorithm based on non-local attention mechanism is proposed.This algorithm used dilated convolution to increase the receptive field,and the non-local attention mechanism pays attention to the feature information of the text to extract useful text information and remove irrelevant information.Experimental results show that compared with the EAST algorithm,this paper's method can effectively reduces the probability of incomplete detection or misdetection.(2)In terms of text recognition,for the problem that CTC and sequence-to-sequence have low accuracy on irregular text and Chinese text,a text recognition model based on hybrid two dimension CTC and attention sequences is proposed.this algor used two dimension CTC to adaptively focus on the spatial location information of the text,and excluding the influence of background noise.Through experimental verification,compared with the CTC and sequence to sequence models,the method in this paper improved the accuracy in irregular text recognition and Chinese text recognition,and accelerated the convergence speed.(3)Based on the natural scene text detection and recognition algorithm proposed in this paper,Python is used as the development language,pycharm is used as the development tool,and Django is used as the web framework to design and implement the natural scene text detection and recognition system.This system function modules: text detection,text recognition and so on.