Computer Vision: Color Spaces, Image Kernels, and Filtering

Vocabulary flashcards covering HSV and RGB color spaces, image convolution kernels, padding, stride, and blurring techniques for computer vision workflows.

HSV Color Space

A color model that separates color type (Hue) from intensity (Saturation and Value), making it more intuitive and effective than RGB for computer vision tasks like object detection.

Hue

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

Saturation

The intensity of a color, measured on a scale from $0$ (gray) to $1$ (pure color).

Value

The brightness level of a color, ranging from $0$ (black) to $1$ (full brightness).

RGB Normalization

The process of scaling RGB values to the range $[0, 1]$ so that HSV conversion formulas can correctly calculate hue, saturation, and brightness.

Delta

The difference between the maximum and minimum normalized RGB values, used to select the appropriate equation for calculating Hue.

Grayscale Input (in HSV Conversion)

An input where Hue and Saturation evaluate to $0$, leaving only the Value channel with a non-zero magnitude.

Image Kernel

An array that slides across an image with specific values determining how much each surrounding pixel in a window contributes to the new center pixel.

Normalized Mean Kernel

A square kernel array where every entry is set to $1$ divided by the total number of entries (such as each entry being $\frac{1}{9}$ for a $3 \times 3$ kernel), ensuring all values sum to $1$.

Kernel Filtering

An operation where the kernel extracts nearby pixel values in a local window, multiplies each by the corresponding kernel weight, sums the products, and reassigns the result to the center pixel.

Sliding-Window Operation

A method of traversing an image by moving a kernel array across pixel locations according to a specified stride and performing neighborhood calculations at each stop.

Zero-Padding

Adding zero-valued pixels around the borders of an image, which allows edge pixels to be centered within the kernel window without shrinking image dimensions.

Zero-Padding Formula

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

Stride

The number of pixels the sliding window shifts at each step; a stride equal to $1$ means the window moves by $1$ pixel at a time.

Odd Kernel Size

A geometric constraint requiring kernel dimensions to be odd numbers so that there is an exact center pixel with an equal number of surrounding pixels on each side.

Gaussian Blur

A filtering technique that assigns larger weights to pixels near the center of the kernel and smaller weights to pixels farther away, smoothing noise while preserving large structures.

Mean Blur

A blurring technique that weights all pixels in the window equally, often producing blocky artifacts and removing excessive structural detail.

Pre-Extraction Blurring

The practice of applying blur before edge detection or feature extraction to suppress small pixel-level noise variations, making larger shapes easier to analyze.

Sigma ($\sigma$)

A Gaussian kernel parameter that controls the spread of the blur; a smaller value produces a tighter blur, whereas a larger value yields a wider blur.