Image Processing Basics - Digital Signal Processing: Fundamentals and Applications - page 711

Finally, we yield the following filtered image:
2
4

3
5

104 105 100 51
101

95 96 54

98 98 102 54

49 54 53 27

As we know, due to zero padding for boundaries, the last-row and last-column values are in error. However, for
a large image, these errors at the boundaries can be neglected without affecting image quality. The first 3 3
elements in the processed image have values that are close to those of the original image. Hence, the image is
enhanced.

Figure 14.25 shows the noisy image and enhanced images using the 3 3, 4 4, 8 8 average
lowpass filter kernels, respectively. The average kernel removes noise. However, it also blurs the
image. When using a large-sized kernel, the quality of the processed image becomes unacceptable.
The sophisticated large-size kernels are used for noise filtering. Although it is beyond the scope of
the text, the Gaussian filter kernel with a standard deviation s ¼ 0 : 9, for instance, is given by the
following:

FIGURE 14.25

Noise filtering using the lowpass average kernels.