Analysis of PCA Usage to Detect and Correct Skew in Document Images

Abstract

In recent years, all the documents are digitized by using different tools like camera and scanner. During the digitization, many mistakes are taken place, such as blur image, noise, and skew formation on entire document. These mistakes are directly influence on the consistency and productivity of the document image segmentation and analysis. In these mistakes, skew detection and correction are important part, and for this many approaches are introduced by different researchers. In this paper, principal component analysis (PCA) procedure is used to find the skew angle on full page. First convert the input document into binary form image, then use Sobel and Gaussian filters to find edges and suppressing the noise, respectively. Then apply the PCA to find the skew angle exists in a document. The PCA procedure first finds the covariance matrix, then generates the eigen values and eigen vectors, and from this calculate the unit vector for principal component. Later using principal component, identify relevant skew angle of the document. To correct the skew, rotate the test sample by preferred angle, and it can remove the skew occurred in document while acquiring the image. The evaluation of system methodology is carried out for different documents; the experimental results are significantly better.