| With the rapid development of information technology, due to the growingdemand for digital office, digital image processing is gradually replacing the manualprocessing of paper documents. Document image is the output that a paper documentis put into computer system through the scanner, digital camera, or documentprocessing system. It has a small storage space, easy management, quick handlingcharacteristics. Digital document image increasingly applied to net view, digitallibraries, banks, records management and other work. Therefore, document imageprocessing has very realistic and practical research value.Document image processing mainly involves two aspects: First, in order toimprove the quality of document image, and improve the accuracy of the documentimage post-processing to optimize the quality of the original document image, such asde-noising, de-skew， etc. The second processing is the image encryption,watermarking, etc. to meet the special needs of the document image. This paperstudies the content of the first aspect, doing a lot of the detailed introduction andresearch of the skew detection and the real-time de-skew of the document image, andmake a series of relevant processing algorithms.Due to the influence of anthropogenic factors and scanner mechanical errors,document image usually has some inclination angle. Document image processing,which is about document layout analysis, understanding, and OCR recognition ofsubsequent processing such as image restoration work, is of great significance, andthe estimation accuracy of tilt angle is particularly important. In large-scale documentimaging procedures, real-time image correction function is often required, so thestudy of high speed image interpolation method and text localization algorithm isnecessary.This paper presents a scan algorithm based on edge features for complexdocument image skew angle estimation in the third chapter. Boundary feature pointdetection method marks angled lines feature points first, then calculate the anglebetween these points and distribution based on the voting method for estimation tiltedangle, and Gaussian wavelet transform transient signal detection method is used forestimating the angle of tilt of the line segment. Experiments show that this method isnot sensitive to noise, and with very high accuracy and stability. In correcting images,this paper proposes a method of high speed interpolation for handling documentimages which usually containing empty area characteristic in the fourth chapter. Withblock statistics analysis of document images feature, the copy method of blank imageblock and bi-linear interpolation method are used in correcting images. This paperpresents the implementation process of the algorithm. Theoretical and experimentalresults show that our algorithm can fully ensure the correction of the scanned image quality, and efficiency is higher than bilinear interpolation algorithm in documentimage scanning and achieved good results. In text line detection, presents a firstderivation step characteristics analysis based on Gaussian function detection of linesof text. First strip project function is used for estimating the density of documentimage, and then applies transient signal analysis method to get a line of text boundarypoints, which form the points of a text line and has the characteristics of continuous,single pixel. Due to the statistical properties of the projection functions and Gaussianfunction of transient signal detection capabilities, which include some noise detectionalgorithm, overlapping the text, line noise distortion insensitive. Experimental resultsshow that it has a very high detection accuracy and stability. |
