3.1 Smoothing Noise with Low-Pass Filters
The first stage in image processing is smoothing noise that can have its origin either in the
testing technique [2] or in the digitization process, the latter known as electronic noise. In this
stage low-pass filters are normaly applied, since noise is characterized by high frequencies
[7,8]. The most widely used filter for smoothing noise in radiographs is the median type. This
filter, the middle value among the values (elements) of the original operation, will be the
output value of the pixel. The median filter has good performance in noise elimination,
without removing relevant defects and without decreasing image sharpness as evidently as
mean and Gaussian filters [8, 10, 11, 20].
Wang [18] points out the difficulty that there is in smoothing noise in radiographs, due to the
variation of gray levels existing in the images, making very difficult the choice of a filter or a
smoothing technique that can be used on all radiographic images. However, he states that the
median filter with 3x3 and 5x5 operating windows is the most adequate to be applied to
radiographs of continuous welds [18].