Segmentation Algorithm Research For Nonuniform Illumination Document Image

Nowdays how to find out the interesting part from the image,that is the image analysis, has become the major research field of digital image process for the scholars while it used to improve the image quality mainly in early time.In order to highlight the part of the content of image features, the common way is the binarization technology; As the key processing step of the program of the OCR, Binarization not only could filter the useless information and highlight useful information, but also directly affect the characters recognition rate in later.When the image’s target and background are simple, it is used global threshold segmentation algorithm could obtain satisfactory effect; However, the image tend to be affected by the interference of non-uniform illumination which impressed the binarization result in practice applications. Aiming at solving the problem, the paper analyzes and researches the cause of non-uniform illumination, and introduces the four kinds of classic local threshold segmentation algorithm at the same time. When the paper introduces the Niblack algorithm, for solving the problem that the algorithm relies heavily on the threshold parameter k and w, it proposed an improved algorithm based on Niblack algorithm. At first, the algorithm define the parameter k as a correction coefficient that changes adaptively with window while the former k is global effective. And the correction coefficient k is substituted into the original Niblack algorithm formula. Then further decrease threshold value appropriately so that it could enhance the ability of antiinterference in the background region. The experiments show that this algorithm’s visual effect is close to Sauvola algorithm.Through contrast experimental analysis, the paper found that the traditional edge operator, global threshold algorithm and local threshold algorithm can’t achieve the both satisfactory results on segmentation effect and algorithm efficiency. Therefore, the paper proposes a fast improved algorithm based on local mean. At first, the algorithm uses integral image fast algorithm to calculate the local mean. Then substitute it into the improved threshold model algorithm. At least complete the segmentation so that further improved effect of image segmentation in non-uniform illumination.Finally, the paper introduces the common method of image quality evaluation. By the comparing experimental results, the paper analyzes and illustrates the improved binarization algorithm improves the calculation efficiency and enhances the image segmentation effect, having the advantage of both on visual effect and processing time.