Data Form Automatic Identification System

The financial department is an indispensable part of modern enterprises. The growth and success of enterprises depend on good financial budget and management. But the most important thing of financial management is financial data management. Nowadays, in order to survive in this fiercely dog-eat-dog world, the successful enterprises must have experienced many years of development. For many years of operation, the enterprises already have accumulated mountains of financial data. Many financial data can show information about past transactions, so it is too important to lose. However, stored in the company, these paper documents not only occupy large space but also tend to be damaged and lost for years of storage. Then the best solution is that these contents of old paper documents must be stored in the computer. In this thesis, the main object of the system is that the data recorded on old paper documents are then placed in a computer in order to facilitate future modification and search financial data.This thesis refers to relative image recognition technologies in the area of financial data identification form in all over the world at the present. After analyzing the main algorithms of preprocessing, shadow elimination, form recognition and so on of financial data form, this system is designed. The recognition of financial statements form is realized finally by obtaining the remarkable results.Firstly, scan or take photos of the paper financial data so as to keep the paper documents in the computer in the form of images, and then analyze the image denoising to get the conclusion that the noise of table layout is mainly salt and pepper noise. The next is a brief introduction about the salt and pepper noise. This paper has made a comparative research into several image denoising methods of salt and pepper noise. The experimental results show that median filtering can reach more satisfactory results to remove noise. Then by the binarization, it is very convenient to recognize financial data form images. During the image acquisition, shadow can occur easily in the data and form at the margin of image. So we need to deal with the shaded area of the form before recognition. The image denoising, binarization and shadow removal will make financial data images to be distinguishable, which is convenient to make the next recognition.Secondly, adjust financial data form images. Based on the fact that the recognition system has a high requirement for image state, the tilted financial data form images tend to have a great influence on recognition results. So this thesis has taken steps for image tilt elimination before recognition. In order to display the form images more normally, this thesis has used the projection method to detect images, and then look for the dip angle to adjust by rotating.Finally, search the form lines of financial data form images. After extracting the form of images, the form lines tend to be broken. What's more, we need to restore and connect the broken form lines, and then combine the finished horizontal lines and vertical lines. In the part of extracting form, the layout of form has been analyzed. The intersection points of horizontal lines and vertical lines are defined as feature points. The method of using the feature points to extract form has been applied to extract the financial data format last.