Computer Vision Checkpoint 1: Color Spaces, Kernel Filtering, and Blurring

Practice vocabulary flashcards covering HSV and RGB color spaces, image kernels, convolution/sliding-window mechanics, padding, and Gaussian vs. mean blur filtering.

RGB

A color model primarily used for display and storage.

HSV

A color model preferred for color manipulation and analysis in computer vision tasks like object detection because it separates color (Hue) from intensity (Saturation and Value) and is more intuitive than RGB.

Hue

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

Saturation

The intensity of a color ranging from $0$ (gray) to $1$ (pure color).

Value

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

RGB Normalization

Scaling RGB values to the range $[0-1]$ to place them in the correct range for HSV conversion formulas, allowing hue, saturation, and brightness to be calculated correctly.

Floating-point Return Values (in rgb to hsv)

Output values used instead of uint8 to prevent decimals from being truncated or rounded too early, preserving accuracy before being cast to integers in convert_to_hsv.

Min and Max (in rgb to hsv)

The minimum and maximum normalized RGB values within a pixel, used to select the appropriate equation for calculating Hue.

Delta

The difference between the maximum and minimum normalized RGB values, which determines the specific equation used to compute Hue (such as when delta is $0$).

convert_to_hsv Function

The function that applies rgb to hsv to every pixel by looping through each pixel in the image and casting floating-point values to uint8 integers.

Image Padding (Task 1)

The addition of pixels to the outside of an image to match the standard size of $100 \times 100$ while preserving the aspect ratio.

Standard Size Resizing

Resizing and padding an image to a consistent dimension of $100 \times 100$ to maintain a uniform size across images.

Grayscale Input Behavior (in convert to hsv)

An input condition where Hue and Saturation evaluate to $0$, while only Value has a non-zero value, still returning valid HSV values.

Image Kernel

An array that slides across an image containing specific values that decide how much each pixel in that window contributes to the center pixel.

Normalized Mean Kernel

A square array where each entry is assigned the value of $1$ divided by the total number of entries (e.g., $1/9$ for a $3 \times 3$ kernel) so that all entries sum up to $1$.

Kernel Filtering

An operation where the kernel extracts pixel values within a local window, multiplies them by the corresponding kernel values, sums them together, and re-assigns the result to the center pixel.

Sliding-Window Operation

An operation that slides a window matching the kernel dimensions across pixels, performs calculations at each location, and advances to the next pixel according to the stride.

Edge Kerneling Without Padding

A condition where edge pixels cannot be processed because the kernel cannot read values beyond the image borders, forcing the output image size to decrease.

Zero-Padding Formula

The method to calculate required padding: subtract $1$ from the kernel size and divide by $2$ ($\frac{\text{kernel size} - 1}{2}$), supplying sufficient boundary values to preserve image size.

Stride

The number of pixels by which the sliding window shifts during filtering (where a stride of $1$ means shifting by $1$ pixel).

Odd Kernel Size Rule

The requirement that a kernel dimension must be an odd number so that there is an exact center pixel and an equal number of surrounding pixels on each side for calculation.

Gaussian Blur Weighting

A weighting distribution that assigns larger weights to values near the center of the kernel and smaller weights to values farther away.

Pre-filtering Blurring

A step performed prior to feature extraction (such as edge detection) to suppress small pixel-level variations like noise, making larger shapes easier to analyze.

Mean Blur Artifacts

Blocky visual artifacts and loss of major image structure caused by applying uniform, heavy weighting across the entire kernel.

Sigma ($\sigma$)

The parameter that controls the spread of a Gaussian kernel, where a smaller sigma creates a tighter blur and a larger sigma creates a wider blur.