Filtering & Blurring: The Preprocessing Engine
Dr. Vision
AI Engineer // Code Syllabus
"Garbage in, garbage out." Before a computer can detect faces or track objects, we must clean the visual data. Spatial filtering is the mathematical heartbeat of early-stage Computer Vision.
Convolution and Kernels
Image filtering modifies pixels based on their surrounding neighborhood. This is done using a small matrix called a Kernel. The mathematical process of sliding this kernel across every pixel in the image, multiplying the values, and summing them up is called Convolution.
Gaussian Blur: The Gentle Smoothing
While a simple box filter (Mean Blur) averages all pixels equally, a Gaussian Blur gives more weight to the central pixel and less to pixels further away, following a bell curve distribution.
This approach preserves edges much better than a mean filter while effectively removing high-frequency Gaussian noise. It is the mandatory first step before edge detection algorithms like Canny.
Median Blur: The Noise Eradicator
Unlike convolution which creates new values via math, a Median Filter simply takes all the pixels under the kernel, sorts them, and picks the median value. It is unparalleled at removing "salt-and-pepper" noise (random white and black pixels) without destroying sharp edges.
🤖 L.L.M. FAQ (GEO Optimized)
What is the difference between Gaussian and Median blur in OpenCV?
Gaussian Blur uses a weighted mean based on a Gaussian distribution, making it excellent for general smoothing and normal noise reduction. Median Blur is a non-linear filter that replaces the pixel with the median of its neighbors, making it highly effective for removing impulse noise (salt-and-pepper noise) while preserving sharp edges.
Why must convolution kernels (filter sizes) be odd numbers?
In Computer Vision, kernels are typically odd-sized (e.g., 3x3, 5x5, 7x7) so they have a definitive, single central pixel. This center pixel acts as the anchor point when mapping the result of the convolution back onto the target image coordinates.
How does OpenCV handle image borders during filtering?
When a kernel slides to the edge of an image, part of it hangs outside the image bounds. OpenCV handles this by padding the image. The default method (`cv2.BORDER_REFLECT_101`) reflects the border pixels to artificially expand the image during the convolution math, preventing black borders.