Research On Fuzzy Prosecution Text Detection And Recognition Methods

China’s information technology construction is in a period of vigorous development,which is an inevitable trend of global technology and information technology development.In this era,procuratorial work must also keep up with the trend and strengthen the construction of information technology in order to improve efficiency and quality.In the era of informationization of prosecution work,most of the documents handed over in the process of inter-departmental cooperation in handling cases are mainly paper documents,so they need to be processed electronically with high-resolution cameras.Due to the influence of various external factors(lighting,stains and distortions,etc.)during the shooting process,resulting in a series of blurred images,extracting effective information from blurred inspection images is a new theoretical and technical challenge.This paper investigates the recognition of text in fuzzy prosecution images,which will be explored in detail through the following aspects.1.The MSER algorithm based on edge enhancement is designed for the problem of poor text detection of fuzzy inspection images by existing algorithms,after extracting the color components in the RGB color space for adaptive median filtering,and MSER+ and MSER-are used to detect the text candidate regions,and the final text candidate region is obtained by merging the detected text candidate regions for each color component.2.The method of adaptive weight fusion of multiple features based on immunogenetic optimization SVM is designed for the problem of non-text regions in the extracted text candidate regions.Firstly,we use the geometric characteristics of text to perform coarse filtering of non-text regions,and then use the HOG,ULBP and SW features of the image to fuse the features according to the designed fusion formula to obtain the fusion features,optimize the optimization step of SVM by using immune genetic algorithm.3.For the problem of prosecution text recognition,a CRNN-based prosecution text recognition algorithm is designed,and the Bi GRU network is chosen to replace the original Bi LSTM network in CRNN to improve the computational speed of CRNN and ensure the timeliness of prosecution text recognition.Improve the stability of network training while solving the problem of small sample size of the prosecution text dataset.Finally,APP Inventior2 software is used for the mobile terminal.