Studies Of Some Problem For Document Images Processing

At the era of highly developed information technology, the urgent demand of paperless office made the requirements which transform the data from paper type (Information are more likely be recorded by paper) into electronic type are more and more intense. This is a realm with study and application. The technology its needed include digital image processing, pattern recognition, optical character recognition and so on. Text image processing is an important part of this field. In this paper, we put forward some solutions for several problems of text image processing.As an important tool in digital image processing, Chain-code can describe the regional info. It transfers the two-dimensional image to unidimensional code. It has good performance on image processing and analysis because of its advantage at some aspect.Skew detection and correction of text image is an important step of the optical character recognition (OCR). Although there is a lot of methods of skew document detection, but lots of theirs efficiency are not well. This paper proposed a skew document detect method based on previous work and the characteristics of chain-code. Compare with others method which before us, because of we needn't scan image many times and don't be affected by noise our method is simple and efficient.Marginal noise will appeared in scanned document images because of inaccuracy scanning. The presence of marginal noise will seriously affect the result of optical character recognition. This paper will introduce several frequently-used marginal noise removal algorithm first, and then based on the border tracing and filling algorithm, we propose a new marginal noise removal algorithm based the marginal noise's property which it has large region and its location near the boundary. This method not only can remove the marginal noise with island but also can remove the marginal noise which not connected with the most peripheral pixel of image. Because of new filling algorithm, experiments show that this algorithm is not only feasible and effective but also faster and more efficient.