- Discuss neighborhood operators in Image processing:
i) Non-linear filtering (median filtering)
ii) Bilateral filtering
iii) Guided image filtering
Answer:
Neighborhood Operators in Image Processing
Neighborhood operators in image processing operate over local regions (neighborhoods) of pixels to perform various image transformations. While linear filters perform weighted sums of pixel intensities, non-linear filters can preserve edges and remove noise more effectively in certain scenarios. This section discusses three important non-linear neighborhood operators: Median Filtering, Bilateral Filtering, and Guided Image Filtering, along with supporting figures and equations.
i) Non-linear Filtering (Median Filtering)
Non-linear filters do not satisfy the principle of superposition, unlike linear filters. One widely used non-linear filter is the Median Filter, particularly effective against impulse noise (shot noise).
Median Filtering
The median filter replaces each pixel with the median value of the neighboring pixels. It is resistant to outliers and performs better than Gaussian filters when noise contains large intensity deviations.
Median filtering selects the median pixel value (Figure 3.19a), making it robust to extreme values in the neighborhood.
ii) Bilateral Filtering
Bilateral filtering combines spatial proximity with pixel intensity similarity to perform edge-preserving smoothing. It was first introduced to computer vision by Tomasi and Manduchi (1998).
This filtering preserves sharp edges while removing noise, unlike Gaussian filters that blur edges.
iii) Guided Image Filtering
Guided filtering enhances a noisy image using a guide image, possibly a flash or edge-enhanced image.
