Font Size: a A A

Geometric Correction Of Multi-view Document Image Based On Smart Phone

Posted on:2020-07-29Degree:MasterType:Thesis
Country:ChinaCandidate:C W ShuFull Text:PDF
GTID:2428330590995479Subject:Software engineering
Abstract/Summary:PDF Full Text Request
Traditional OCR technology uses scanners or cameras to acquire document images,and then uses computers to process document images to recognize the text on them.Traditional OCR has two main problems besides recognition algorithm,namely,document page distortion and poor device portability.The former will reduce the recognition accuracy of OCR,while the latter will greatly reduce its suitable occasion.In this paper,a multi-view document image geometric correction technology based on smartphone is proposed,it is presented to use the computing and photographing capabilities of smartphones which have been greatly improved in recent years to achieve document image acquisition and geometric correction on smartphones,thereby solving the aforementioned problems.The main work of this paper is to establish a multi-view document image processing framework on smartphone.The processing framework includes multi-view document image acquisition,document image feature point matching and mismatching elimination,document page 3D reconstruction based on SfM,geometric correction based on developable surface model,document image character recognition.In this framework,feature point matching is the key link,and its accuracy seriously affects the system performance.Based on its importance,an ORB-oriented mismatching feature points elimination algorithm is proposed in this paper.The similarity of feature points can be obtained by calculating the angle cosine between feature descriptors,and the matching pairs with less similarity can be eliminated to improve the accuracy of feature point matching.The second important part of the framework is model reconstruction and correction.Based on the idea of Structural From Motion,the coordinates of 3D point cloud are solved using matching point pairs.Considering that the developable surface model covers almost all warping documents,a curve-to-surface fitting method is proposed to reconstruct the developable surface model of the document,and the surface function is used to correct the distorted document image.Finally,a mobile app for geometric correction of multi-view document images is established based on the above research results.The validity of the document image geometric correction algorithm is verified by experiments in this paper,the document image correction app implemented on Android smartphone also has strong practicability.
Keywords/Search Tags:Optical Character Recognition, Smartphone, Document Image Geometric Correction, Feature Matching, SfM, Developable Surface Model
PDF Full Text Request
Related items