Filtering a noisy hand written image - Image Processing Techniques
Version number:

BuyBuy Now DownloadDownload

Pricing Feature Chart Online Manual ActiveX Online Manual

Filtering a noisy scanned hand written image

The original image is a scanned, hand written image with a strong additive noise due to the scanning process and the poor quality of the original paper. Images such as the one used in this example can be filtered in order to remove the noise from it.

It is important to note that simple filtering alone cannot create wonders, and it’s also important to bear in mind that none of the filtering methods are able to make a new image containing more information than the original. After filtering we always lose some information, but if the filtering is chosen well then the information that we need stays on the image, and we lose the noise. A filtering method is only able to create a new image where the information important to us appears stronger than before the filtering.

One of the best filters (in these special cases like this example) for removing additive, salt and pepper noise is a very simple filtering method called the Median filter.

Using the Median filter, the additive salt and pepper noise is removed, but we also lost some information from the handwriting itself. In this case a 2x2 sized matrix was used for the Median filtering due to the small size of the noise regions. The larger the matrix size, the more noise is removed, but also more information is lost from the important parts of the image. So the best solution is to find a middle ground that removes the most noise while preserving the most image integrity.

In the next step, the errors in the hand writing generated by the Median filter should be fixed. This image contains only black and white pixels. There are many excelent filtering methods for these binary images as they are called, but unfortunately the original images are not always binary. So the image should be binarized now.

For binarization, the Floyd-Steinberg error diffusion dithering method was used. On the already binarized image the holes in the hand writing due to the Median filter could be filled up with a binary Dilation filter. Dilation filter makes the lines stronger, fills up the small holes and make the lines continuous. In this example a 3x3 sized full one matrix was used for dilation.

The final filtered image after dilation (above) and the original image (below)