Computer Vision: Color Spaces and Kernel Filtering

Vocabulary flashcards covering HSV color representation, RGB-to-HSV conversion, kernel convolution, image padding, and Gaussian vs. mean blur filtering.

HSV Color Space

A color model commonly preferred in computer vision for color manipulation and analysis because it separates color (Hue) from intensity (Saturation and Value).

Hue

The color type represented by an angle from $0$ to $360$ on a color wheel.

Saturation

The intensity of color in the HSV color model, ranging from $0$ (gray) to $1$ (pure color).

Value

The brightness component of the HSV model, ranging from $0$ (black) to $1$ (full brightness).

RGB Normalization

Scaling RGB channel values to the range $[0, 1]$ so they match the expected input range of HSV conversion formulas.

Delta (in RGB to HSV Conversion)

The difference between the maximum and minimum normalized RGB values ($\text{max} - \text{min}$), used to determine which equation to apply when computing Hue.

Grayscale Input in HSV

An input condition where Hue and Saturation evaluate to $0$, leaving only the Value channel to represent pixel intensity.

Image Kernel

A square array of numerical coefficients that slides across an image to determine how neighboring pixels contribute to the new value of the center pixel.

Normalized Mean Kernel

A square kernel where each element is equal to $1$ divided by the total number of kernel entries, ensuring that all weights sum to $1$.

Sliding-Window Operation

The process of positioning a kernel over a local image window, calculating the sum of element-wise products, and shifting to subsequent pixels according to the stride.

Image Padding

Adding extra pixel borders around an image so edge pixels can be processed by a kernel without shrinking the output dimensions.

Zero-Padding Formula

The formula used to determine the required border padding width: $\frac{\text{kernel size} - 1}{2}$.

Stride

The step size by which the kernel shifts across the image (e.g., a stride of $1$ means shifting by $1$ pixel).

Odd Kernel Size Requirement

The convention of using an odd dimension (such as $3 \times 3$) so that the kernel possesses an unambiguous center pixel surrounded by an equal number of pixels on all sides.

Gaussian Blur Weighting

A filtering approach that assigns larger weights to pixels near the center of the kernel and progressively smaller weights to pixels farther away.

Pre-Blurring

Applying a smoothing filter before edge detection or feature extraction to suppress small pixel-level noise variations, making dominant structures easier to detect.

Mean Blur Artifacts

A blocky visual behavior and excessive loss of edge structure that occurs because mean filtering applies equal weight to all neighboring pixels.

Sigma ($\sigma$)

The parameter that controls the spread of a Gaussian kernel, where smaller values yield a tighter blur and larger values produce a wider blur.