Introduction to Image Processing and Computer Vision - Practice Assignment 3

Comprehensive vocabulary flashcards covering image filtering, analysis techniques, filters (linear and non-linear), pyramids, and computational efficiency based on Lesson 5 and 6 lecture notes.

Image filtering

The process of transforming an image into a new image by applying a rule that changes pixel values in a controlled way.

Sliding-kernel

A local neighborhood that moves across the image during filtering.

Box filter

A filter that replaces each pixel with the average of its neighborhood, causing the image to smooth or blur.

Gaussian filter

A filter that gives more weight to pixels near the center of the window, producing a more natural blur with fewer blocky artifacts than a box filter.

Separable filter

A 2D filter that can be implemented as two 1D convolutions (one using a column filter and one using a row filter), reducing computational cost.

Convolution

An operation where the kernel is flipped horizontally and vertically before being applied to the image.

Cross-correlation

An operation that computes local weighted sums between the image and a kernel without flipping the kernel.

Linear shift-invariant (LSI) filtering

A type of filtering where each output pixel is a linear combination of neighboring input pixels and the same kernel is used at every location.

Sobel filter

A filter used to estimate image gradients and detect edges by identifying sudden intensity changes.

Median filter

A non-linear filter that is especially effective for removing salt-and-pepper noise.

Image pyramid

A collection of images of the same scene captured at different resolutions.

Aliasing

False patterns or distortion that occur when an image is subsampled without prior filtering to remove high-frequency components.

Gaussian pyramid

A collection of progressively smaller and smoother versions of an image, where the next level typically has a size of $\frac{M}{2} \times \frac{N}{2}$.

Laplacian pyramid

A pyramid that stores the residual details or the difference between Gaussian pyramid levels to make reconstruction of the original image possible.

Low-pass filter

A frequency filter mainly used for smoothing or blurring images.

Image gradients

Representations of the directional change in image intensity, strongly related to edges and boundaries.

Derivative of Gaussian (DoG)

A filter that combines Gaussian smoothing to reduce noise with differentiation to highlight intensity transitions.

Laplacian of Gaussian (LoG)

A second derivative operator applied after Gaussian smoothing used to detect edges and blob-like structures.

Dilation

A non-linear operation that expands bright regions by taking the maximum value in the neighborhood.

Erosion

A non-linear operation that shrinks bright regions by taking the minimum value in the neighborhood.

Low frequencies

Image components corresponding to smooth intensity changes and coarse image structure.

High frequencies

Image components corresponding to rapid intensity changes such as edges, textures, and fine details.

Computational cost (Non-separable N x N filter)

The total operations required for an image of size $M \times M$, calculated as $M^2 \times N^2$.

Computational cost (Separable filter)

The reduced total operations required, calculated as $2M^2N$, by splitting a 2D convolution into two 1D passes.

Valid output mode

A boundary handling mode where only positions where the kernel fully fits inside the image are computed, resulting in a smaller output image.