64
2: Raghuveer M. Rao, Manoj K. Arora
a
b
Fig.2.S. a Input image. b Result after smoothing
Then, the effect is one of smoothing. This operation is most commonly employed for suppressing the effects of noise as demonstrated in Fig. 2.8. In this
instance, the filter (kernel) in (2.11) was employed. However, one may choose
other regions of support and a different set of coefficients. Smoothing is invariably accompanied by a loss of high-frequency information in the image.
Here frequency refers to spatial, not optical, frequency (or gray levels).
The operation in (2.9) is a linear convolution in two dimensions regardless
of the extent of support of A. If A is not only finite but also small, the sequence
h(k, e ), also known as the convolution kernel, is represented as a matrix, with k
indexing the rows and £ indexing the columns. For example, the uniform 3 x 3
smoothing kernel becomes
h(k, £) = [~~~ ~~~ ~~~] .
1/9 1/9 1/9
2.5.2.2
Edge Operations
The human visual system is sensitive to edges (Gonzalez and Woods 1992).
As such edge detection, extraction and enhancement are employed for various purposes including assisting manual interpretation of images, automatic