Research On Document Image Preprocessing Based On EmguCV

In daily life,in order to facilitate the preservation and communication of documents,people often need to digitize paper documents.The way to digitize documents is usually through scanner scanning or camera photography,and the image obtained by this way which with document information.is the document image.At present,the commonly used document image processing software on the market uses the global threshold method for the binarization of document images.And it is appropriate to use such methods for the document images with even illumination,but for the document images with uneven illumination a large black area or white area appears after this procedure.When the document image is acquired,there is often noise due to the acquisition device,the preservation status of the document itself,and the subsequent binarization processing steps.The noise affects the visual perception of the user of the document image.In addition,when the document image is acquired,the angular difference between the document and the acquisition device may cause the captured document image to have a skew deformation,and the obliquely deformed document image may affect the visual perception and the accuracy of the OCR recognition.In recent years,as smartphone manufacturers add camera watermarks to photos taken after their mobile phones are taken,the document images captured by mobile phones will also have camera watermarks,and the presence of camera watermarks will interfere with the late OCR recognition accuracy.In response to these problems,a series of pre-processing tasks such as binarization,filtering,tilt correction,and camera watermark removal are required for the document image.The main work is as follows:1)Based on EmguCV(Emgu Computer Vision),a document image preprocessing software was developed using C#,which can perform binarization,filtering,tilt correction and camera watermark removal on the document image.2)In the developed document image pre-processing software,when binarizing a document image with uneven illumination,the Niblack binarization method can avoid large black or white regions existing when processing using the Otsu method.In general,the binarization effect achieved by the developed document image preprocessing software is superior to the binarization of document image processing software commonly used in the market.3)In the developed document image preprocessing software,a filtering function is provided,which can process the noise of the document image,especially through morphological filtering,which can process the noise generated by the binarization of the Niblack method.4)For the tilt correction of the document image,the text line circumscribing rectangle is used to obtain the tilt angle for correction.The method first binarizes the document image,then performs color inversion,and then obtains a rectangular strip of the text line by means of a morphological operation,thereby obtaining a circumscribed rectangle of all the text line strips,and obtaining the longest circumscribed rectangle.The tilt angle of the circumscribed rectangle is obtained to rotate the document image.The correction effect of the method can meet the requirements of the image image tilt correction.The tilt correction for the Chinese document image is basically consistent due to the Chinese upper and lower heights,and the effect ratio is based on the least squares method.The tilt correction is better.5)For the camera watermark existing in the document image,firstly analyze the characteristics of the camera watermark,and obtain the optimal threshold by comparing the binarization effects under different thresholds,then binarize the camera watermark,and finally use the common The image restoration algorithm-FMM algorithm removes the camera watermark,and the removal effect is very good,and basically does not cause character interference to OCR recognition.