Text images acquired through the use of a scanner can range in quality from excellent to poor, depending on a number of factors. Insufficient scanning resolution, other non-ideal scanning parameters or a low-quality hard copy can all lead to image files with text that is difficult to read. A number of steps can be taken to resolve such difficulties and to improve the readability of scanned text.
The human mind has a remarkable ability to recognize patterns in the images transduced by the eye. As such, even jumbled or highly distorted text can be read with incredible accuracy. Although a simple problem for people, pattern recognition is a notoriously difficult problem for computers, and the software solutions available for this problem are often plagued with drawbacks. Nevertheless, despite shortcomings in this area, pattern recognition can be extremely useful in the area of scanned text. If the original hard copy document being scanned is highly damaged or stained, either through heavy use or through abuse, there may be little that can be done to recover the text. Nevertheless, through various image processing and pattern recognition techniques, in many cases, there is a strong possibility of vastly improving the readability of the text.
The process of converting text in an image into a discrete set of characters that is editable in a word processor or other text editing software package is called optical character recognition (OCR). In some cases, text that is sufficiently clear can be input directly into image processing software with OCR capabilities, thus yielding an editable document containing the text. Since the text has been identified independently of the scanned image, the readability is improved almost infinitely. For text that has poor readability, some pre-processing may be required prior to the use of an OCR routine. For example, the image containing the text may have a very dark background or low-contrast resolution of the text characters. In such a case, the use of the gamma correction and contrast adjustment tools in an image processing package will often be a tremendous help. It may also be beneficial, if the scanned image is in color, to convert to a gray scale or black and white image. Sharpening of the image can increase the distinctness of the characters as well.
The background of a scanned portion of text is brightened and the contrast is increased.
Once processed using an OCR routine, the text image becomes a set of characters that can be clearly read or edited in a word processing program. Again, however, OCR software is based on a pattern recognition algorithm; the results, therefore, must be checked for errors, since pattern recognition by computers may periodically produce incorrect results. It is crucial, whatever the case, to understand the requirements and limitations of the particular OCR software being employed, so as to achieve the best results. To be sure, the more complicated (and expensive) packages will usually perform more accurately, but a little user preprocessing of the image can improve the performance of virtually any OCR program.
Even if the scanned text is not to be converted using an OCR routine, an image processing package can still provide significant improvements to blurred or distorted text. An array of tools for adjusting the sharpness, contrast, brightness and other aspects of the image, along with tools for eliminating various undesirable image artifacts, can be extremely beneficial for cases involving poor readability of scanned texts.
Author: Jeffrey Clark