A Theoretical Comparison of Skew Detection Techniques

Transcription

1 A Theoretical Comparison of Skew Detection Techniques Er.Jyoti Rani Assistant Professor,Giani Zail Singh College of Engineering And Technology,Bathinda,Punjab,India. Neha Watts Student, Giani Zail Singh College of Engineering And Technology,Bathinda,Punjab,India. Abstract This paper represents different methodologies for the text skew estimation in binary images/gray scale images. They have been used widely for the skew identification of the printed text. There are so many ways for detecting and correcting a slant or skew in a given document or page. Some of them provides us accuracy but are slow in speed, and some of them are good for only small text but fails in the case of multiline and multicolumn text or others have angle limitation drawback. In this paper it ends up with limitation of the existing techniques.. Keywords Document image processing, Skew detection, Skew correction, FFT. I. INTRODUCTION The field of image processing brings out the idea of automatic gathering and processing of the observed information. Like, most of the documents are present in the printed form. But if they are needed to be converted into electronic form, it has to be done through scanning. During document scanning [1] [3], skew is inevitably introduced into the incoming document image. Skew refers to the text which neither parallel nor at right angles to a specified or implied line. Character recognition is very sensitive to the page skew, skew detection and correction in document images are the critical steps before layout analysis. The different types of skews within a document page can fall into these categories:- Global Skew, assuming that all page text have the same angle skew, Multiple-Skew, when certain area of the page have different slant than other, and Non- Uniform text line skew, when the slant fluctuates (such as lines having a wavy shape). Skew detection is used for text line position determination in Digitized documents, automated page orientation, and skew angle detection for binary document images, skew detection in handwritten scripts, in compensation for Internet audio applications and in the correction of scanned documents. The largest classes of methods for skew detection are based on projection profile analysis, Hough transform, nearest-neighbor clustering, fourier- transformation, histogram analysis, moments and other methods. During document scanning, skew is inevitably introduced into the incoming document image. Skew detection is one the first operations to be applied to scanned documents when converting data to a digital format. Its aim is to align an image before processing because text segmentation and recognition methods require properly aligned next lines. Most of the times, document analysis systems require prior skew detection and correction before the images are forwarded for character recognition, layout analysis, document image mosaicing and many other applications. An efficient and accurate method for determining document image skew is therefore an essential need. A number of methods have previously been proposed for identifying document image skew angles. However, most of the IJCSIET-ISSUE3-VOLUME2-SERIES3 Page 1

2 approaches are designed mainly for machine printed documents.most of the times, document analysis systems require prior skew detection and correction before the images are forwarded for character recognition, layout analysis, document image mosaicing and many other applications. An efficient and accurate method for determining document image skew is therefore an essential need. A number of methods have previously been proposed for identifying document image skew angles. However, most of the approaches are designed mainly for machine printed documents. Many approaches of skew detection can process pure textual document images successfully. But it is a challenging problem to process documents with large areas of non-textual contents. II. RELATED WORK Nearest-neighbour based approach [1], [2]. [1] has high accuracy but with size restriction drawback.the chains with a largest possible numbers of N-N pairs are selected, and their slopes are computed to give the skew angle of document images. This can detect angles of different documents without the skew angle limitation and without the requirement of predominant area. It is also able to deal with different scripts and even different text orientations appearing on the same image. It is able to deal with documents of different scripts such as English, Tamil and Chinese. [2] detects the global skew angle angle of document images within the range [-90, 90] with a generic, scale-independent technique. By using the same framework, the algorithm is then extended for Roman script documents so as to cope with the full range [-180, 180]. The improved version is very fast and requires no explicit parameters. They offer the greatest flexibility, accuracy and run-time performance. Skew detection using Moments [3], [4], [5] [3] finds the primary axis of every object in the document rather than the Hough transform and handles arbitrary angles. The object can be a halftone image or a text line. Lines are described using the chain code. This works in any kind of documents even when they contain images mixed with text. Both multiple line skew and global skew can be detected, but the angle must be in the range of -30 degrees to +30 degrees. This method can be used in Optical Character recognition. [4] Connects the bounding boxes and inside the longest connected component, the gravity centres are determined for each bounding box. They represent the reference points, which are used for the calculation of the initial skew rate. The comparative analysis of the origin and estimated skew rate is used to evaluate. It is robust for the skew estimation in the wide range of resolutions. It is suitable for multiline and multi-column without any segmentation. Hough transformation approach [5], [6], [7] [5]The non-textual data in a binary image constitutes the main source for the page orientation classification and the document skew angle errors. By dividing a binary image into squares in the page orientation case and using the bottom pixels of candidate objects in the skew angle case, our system of algorithms is able to reduce the inclusion of non-textual data and thus to increase the detection accuracy rate The performance of the Hough transform used to detect the skew angle is sped up using (1) data reduction procedure, which simplifies a chosen square by preserving only the bottom pixels of candidate objects; and (2) page orientation information to detect skew angle on a IJCSIET-ISSUE3-VOLUME2-SERIES3 Page 2

3 chosen square area instead of the entire binary image. [6] considered some selected characters of the text which may be subjected to thinning and Hough transform to estimate skew angle accurately. This method is superior with respect to accuracy, computational time and suitability. But fail for document images containing text with picture. [7] proposed a fast skew estimation and correction algorithm for English and Korean documents based on a Block Adjacency Graph representation. This method uses a coarse/refine strategy based on Hough transformation of connected components in the image. Speed and accuracy has been achieved as this confine the region of interest to a small and plausible text area, and confine the search range of Hough transform to Q, where Q is a coarse skew angle. Cross-correlation method for skew detection [8], [9] [8] describes a robust yet fast algorithm for skew detection in binary document images. Instead of finding the correlation for the entire image, it is calculated over small regions selected randomly. This does not require prior segmentation of the document into text and graphics regions. The maximum median of cross-correlation is used as the criterion to obtain the skew, and a Monte Carlo sampling technique is chosen to determine the number of regions over which the correlation have to be calculated. This can be applied to different linguistic scripts. [9] Modified cross-correlation method uses horizontal and vertical cross-correlation simultaneously to deal with vertically laid-out text, which is commonly used in Chinese or Japanese documents. Here, small randomly regions are selected to speed up the process. The region verification stage and further processing of auxiliary peaks make the method robust and reliable. Fourier transformation method for skew detection [10], [11] [10] presents a new Averaged Block Directional Spectrum (ABDS).Skew determination is important because many other document analysis techniques require (near) perfectly aligned document images to work well. The technique is based on calculating the average 2D Fourier transform of blocks in a document image and using the Radon transform to find the peak in the directional spectrum. This technique is highly accurate, obtaining the skew angle within 0.25 in 96.4% of test images. The technique has also shown robustness to documents highly corrupted by noise or having a large number of graphical elements. [11] Proposed skew detection algorithm which has no restriction on detectable angle range and does not rely on large blocks of text. It works well on textual document images, graphical images and mixed text and graphic images. The skew detection algorithm is robust when compared with other methods when very few text lines are present in the document image. The Fourier transform method is a representation of the projected profile method in the Fourier domain. The results are mathematically identical, but Fourier transform is only a different approach to the same text and document properties projection profile is based upon [10], [11]. Kumar et al. (2012)[12] has discussed that in optical character recognition and document image analysis skew is introduced in coming documented image. Which degrade the performance of OCR and image analysis system so to detection and correction of skew angle is important step of preprocessing of document analysis. Many methods have been proposed by researchers for the detection of skew in binary image documents. The majority IJCSIET-ISSUE3-VOLUME2-SERIES3 Page 3

4 of them are based on Projection profile, Fourier transform, and cross-correlation, Hough transform, Nearest Neighbor connectivity, linear regression analysis and mathematical morphology. Main advantage of Hough transform[13] is its accuracy and simplicity. But due to slow speed many researchers work on its speed complexity without compromising the accuracy. So, for improving computational efficiency of Hough transform there are various variations have been proposed by researchers to reduce the computational time for skew angle. Kumar et al. (2012)[14] has introduced new method which reduces the time complexity without compromising the accuracy of Hough transform. Singh et al. (2008) [15] has studied that the Hough transform provides a robust technique for skew detection in document images, but suffers from high time complexity which becomes prohibitive for detecting skew in large documents. Analysis of time complexity on various stages of skew detection process is carried out in this paper. A complete skew detection and correction process is divided into three parts: a preprocessing stage using a simplified form of block adjacency graph (BAG), voting process using the Hough transform and de-skewing the image using rotation. Skew correction [16] phase, which is hitherto a neglected area, is analysed for the quality of de-skewed images with respect to the type of rotation. Fast algorithms for all the three stages are presented and exhaustive analysis on time complexity is conducted. It is shown that the overall time taken for the whole process is less than one second even for very large documents. It is also observed that time taken in rotation is as significant as in skew detection which is reduced with the help of fast algorithms using integer operations. While the BAG algorithm is found to be effective for documents with Roman script, it does not provide satisfactory results for Indian scripts where headline is a part of a script. Cao et al. (2003) [17] has demonstrated that during document scanning, skew is inevitably introduced into the incoming document image. Since the algorithms for layout analysis and character recognition are generally very sensitive to the page skew, skew detection and correction in document images are the critical steps before layout analysis. In this paper, a novel skew detection method based on straight-line fitting is proposed. And a concept of eigen-point is introduced. After the relations between the successive eigen-points in every text line within a suitable sub-region were analyzed, the eigen-points most possibly laid on the baselines are selected as samples for the straightline fitting. The average of these baseline directions is computed, which corresponds to the degree of skew of the whole document image. Then a fast skew correction method based on the scanning line model is also presented. Experiments prove that the proposed approaches are fast and accurate. Li et al. (2007) [4] has proposed a novel document skew detection algorithm based on wavelet decompositions and projection profile analysis. First, the skewed document images are decomposed by the wavelet transform. The matrix containing the absolute values of the horizontal subband coefficients, which preserves the text s horizontal structure, is then rotated through a range of angles. A projection profile is computed at each angle, and the angle that maximizes a criterion function is regarded as the skew angle. Experimental results show that this algorithm performs well on document images of various layouts and is also robust to different languages. IJCSIET-ISSUE3-VOLUME2-SERIES3 Page 4

5 The effects of various wavelet basis, number of decomposition levels, and parameters of the criterion function are investigated too. Yin (2001) [5] took a study and proved that the number of daily-received paper-based office documents is overwhelming, the development of document image analysis, which converts the paper-based documents into electronic forms becomes increasingly important. This paper describes a skew detection method which first smoothes the black runs and locates the black white transitions to emphasize the text lines. Then the skew angle is determined by an improved Hough transform. For the block classification step, a rule-based classifier is presented. The classification rules are derived from the gray level entropy, block aspect ratio, and run length analysis. Kapoor et al. (2004) [18] have proposed two algorithms. The first one detects the skewing of words and the second corrects the skewing from handwritten words. Both algorithms make use of the Radon transform based projection profile technique. The method does not require preprocessing and it works equally good even with noise. The method is fast. The algorithms have been tested on words taken from more than 200 writers and the results obtained confirm the overall accuracy of proposed system. No error was detected. Egozi et al. (2011) [19] has presented a statistical approach to skew detection, where the distribution of textual features of document images is modeled as a mixture of straight lines in Gaussian noise. The Expectation Maximization (EM) algorithm is used to estimate the parameters of the statistical model and the estimated skew angle is extracted from the estimated parameters. Experiments demonstrate that our method is favorably comparable to other existing methods in terms of accuracy and efficiency. Shivakumara et al. (2006) [20] has shown that skew angle estimation is an important component of optical character recognition (OCR) systems and document analysis systems (DAS). A novel and an efficient method to estimate the skew angle of a scanned document image is proposed. The proposed method has two stages. In first stage, using boundary-growing approach, text lines containing characters of the scanned document image are extracted. From each text line, coordinates of the positions of the characters are obtained. In second stage, the obtained coordinates are fed to linear regression analysis (LRA) for the purpose of computation of skew angle. Several experiments have been conducted on various types of documents such as documents containing different language texts, documents with different fonts and documents with noise to reveal the robustness of the proposed method. Makridis et al. (2010) [21] has proposed a new technique for global and local skew detection in complex color documents. The proposed technique, which can be applied also to grayscale and binary documents, consists of four main stages; color reduction, text localization, document binarization and skew correction. Color reduction limits the initial number of colors to a small number, usually smaller than 10 colors. Thus, the original documents are decomposed in homogenous regions. Text localization initially divides the document into a number of binary planes (color planes) equal to the number of the reduced colors. Then, connected component analysis is performed and text is extracted according to similarity features between adjacent connected components. In the third stage the binary document is composed by the processed IJCSIET-ISSUE3-VOLUME2-SERIES3 Page 5

6 binary planes. Finally, skew correction is achieved by detecting the direction of connection of the connected components in the binary images. The proposed technique has been extensively tested with two databases, one with complex scanned color cover books and another with grayscale scanned newspapers taken from the database of the university of Oulu. Mascaro et al. (2010) [22] has shown that Skew correction of scanned documents is a crucial step for document recognition systems. Due to the problem of high computational costs of the state-ofthe-art methods, we present herein a variation of a parallelograms covering algorithm. This variation strongly reduces the computational time and works over noisy documents and documents containing non-textual elements, like: stamps, handwritten components and vertical bars. Experimental studies with different databases show that this variation overcomes well-known techniques, achieving better results over synthetic rotated documents and real scanned documents. CONCLUSION There are so many ways for detecting and correcting a slant or skew in a given document or page. Like, Hough transformation, Fouriertransformation, nearest neighbor, cross- correlation, moments, etc. But mostly every technique has some limitations, like some of them provide us speed but are suitable only for small text, some provide us accurate results but are slow in speed, some are costlier if they are good in speed and accuracy. So we are going to propose a new technique with Black and white, or Gray Scale images, which will reduce the time and cost complexity. REFERENCES [1] Y. Lu, C. L. Tan., Improved Nearest Neighbor Based Approach to Accurate Document Skew Estimation, Department of Computer Science, School of Computing, National University of Singapore,Kent Ridge Singapore,2003. [2] K. Iuliu,E. Stefan,S. Christoph., Fast Seamless Skew and Orientation Detection in Document Images, Fraunhofer Institute for Intelligent Analysis and Information Systems (IAIS) Schloss Birlinghoven, Sankt Augustin, Germany,2010. [3] K. George,K. Nicholas., A Fast High Precision Algorithm for the Estimation of Skew Angle Using Moments, Department of Informatics and Telecommunications University of Athens,2002. [4] B. Darko, Dragan R. M., An Algorithm for the Estimation of the Initial Text Skew, University of Belgrade, Technical Faculty Bor, Mining and Metallurgy Institute, Department of Informatics,Serbia, [5] L. Daniel, R.T. George and W. Harry., Automated Page Orientation and Skew Angle Detection for Binary Document Images, Lister Hill National Center for Biomedical Communications, National Library Of Medicine,Department of Computer Science, George Mason University, Fairfax,USA, [6] A. Manjunath, G. Hemantha, and P. Shivakumara., Skew Detection technique for Binary Document Images based on Hough Transform, [7] H. Kwag, S. H. Kim, G. S. Lee., Efficient skew estimation and correction algorithm for document images, Department of IJCSIET-ISSUE3-VOLUME2-SERIES3 Page 6

7 Computer Science, Chonnam National University, South- Korea,2002. [8] Chaudhuri. A, Robust detection of skew in document images. Dept. of Electr. Eng., Indian Inst. of Technol., Bombay, India, [9] Chen, Ming, A robust skew detection algorithm for gray scale document image, Dept. of Electron. Eng., Tsinghua Univ., Beijing, China,1999. [10] AMIN, A., WU, S. Robust skew detection in mixed text/graphics documents.in Internat. Conf. on Document Analysis and Recognition - ICDAR. Seoul (Korea), [11] LOWTHER, S., CHANDRAN, V., SRIDHARAN, S. An accurate method for skew determination in document images. In Proceedings of the Digital Image Computing Techniques and Applications. Melbourne (Australia), [12] Deepak Kumar, Dalwinder Singh, "Modified Approach of Hough Transform for Skew Detection and Correction in Documented Images". International Journal of Research in Computer Science, 2 (3): pp , April [13] Chandan Singh, Nitin Bhatia, Amandeep Kaur, Hough transform based fast skew detection and accurate skew correction methods, Pattern Recognition, Volume 41, Issue 12, December 2008, Pages [14] Yang Cao, Shuhua Wang, Heng Li Skew detection and correction in document images based on straight-line fitting, Pattern Recognition Letters, Volume 24, Issue 12, August 2003, Pages [15] Shutao Li, Qinghua Shen, Jun Sun Skew detection using wavelet decomposition and projection profile analysis, Pattern Recognition Letters, Volume 28, Issue 5, 1 April 2007, Pages [16] P.-Y Yin Skew detection and block classification of printed documents, Image and Vision Computing, Volume 19, Issue 8, 1 May 2001, Pages [17] Rajiv Kapoor, Deepak Bagai, T.S. Kamal A new algorithm for skew detection and correction, Pattern Recognition Letters, Volume 25, Issue 11, August 2004, Pages [18] Amir Egozi, Its hak Dinstein Statistical mixture model for documents skew angle estimation, Pattern Recognition Letters, Volume 32, Issue 14, 15 October 2011, Pages [19] P. Shivakumara, G. Hemantha Kumar A novel boundary growing approach for accurate skew estimation of binary document images, Pattern Recognition Letters, Volume 27, Issue 7, May 2006, Pages [20] Michael Makridis, Nikos Nikolaou, Nikos Papamarkos An adaptive technique for global and local skew correction in color documents, Expert Systems with Applications, Volume 37, Issue 10, October 2010, Pages [21] Angélica A. Mascaro, George D.C. Cavalcanti, Carlos A.B. Mello Fast and robust skew estimation of scanned documents through background area information, Pattern Recognition Letters, Volume 31, Issue 11, 1 August 2010, Pages IJCSIET-ISSUE3-VOLUME2-SERIES3 Page 7