Digital Image Processing and Computer Vision - Practice Flashcards

Flashcards covering the fundamentals of digital images, ISP pipelines, sampling, quantization, and image enhancement techniques like histogram equalization and gamma correction.

A digital image is formed after the measurement, sampling, and __________ of light.

quantization

Modern cameras do not directly output RAW sensor data primarily because RAW is __________ and requires processing to become viewable.

mosaiced

The ISP step responsible for reconstructing missing colors in a Bayer-mosaiced RAW image is __________.

Demosaicing

The main purpose of white balance is to correct the __________ __________ caused by the illumination.

color cast

Sampling in digital imaging refers to measuring a continuous spatial signal at __________ __________ __________.

discrete pixel locations

__________ in digital imaging refers to mapping continuous-valued measurements to discrete intensity levels.

Quantization

If spatial sampling is too low relative to scene detail, the most common result is the appearance of __________ artifacts.

aliasing

Increasing the number of quantization levels while keeping the same sampling typically provides better __________ __________ and smaller quantization error.

intensity precision

A grayscale histogram represents the distribution or frequency of __________ __________ regardless of their spatial location.

intensity values

A point or pixel-wise operation is described as a mapping applied independently to each pixel value using the __________ __________.

same rule

A mapping is __________ __________ if it does not depend on pixel coordinates and the same rule is used everywhere.

spatially invariant

For an 8-bit grayscale image, the valid representable intensity range is $.$

$$[0, 255]$$

After pixel arithmetic, __________ is used to force computed values back into the valid representable intensity range.

clamping

For visualization of frame differencing, the expression __________ is often preferred because it treats both brightening and darkening as change without negative values.

$$|I_t - I_{t-1}|$$

In motion detection, __________ __________ is primarily used to convert a difference image into a binary motion mask.

global thresholding

Robust contrast stretching, also known as percentile clipping, uses $q_{low}$ and $q_{high}$ to reduce the influence of __________ __________ in histogram tails.

extreme outliers

The core idea behind histogram equalization is to use the __________ to remap intensities so output levels are used more uniformly.

CDF

In histogram equalization, contrast is increased in regions where the CDF is __________.

steep

For normalized intensities $a \in [0, 1]$, gamma correction is modeled by the power-law equation __________.

$$b = a^{\gamma}$$

Alpha blending is defined by the equation __________, where $\alpha$ determines the contribution of each image.

$$I_{blend} = \alpha I_{left} + (1 - \alpha)I_{right}$$

A large feathering window can create __________ artifacts if edges or textures in the overlap region are not perfectly aligned.

ghosting

A 2D or joint histogram counts how frequently pairs of values occur together and is more informative than 1D histograms because it captures __________ between variables.

correlation

In the HSV color model, H stands for Hue, S stands for __________, and V stands for Value.

Saturation

Displays use the RGB color model because they are additive systems, whereas printing uses the __________ model because inks absorb light.

CMYK

The YCbCr model is useful because it separates __________ from chrominance, allowing contrast enhancement without affecting color stability.

luminance

The ISP correction step that subtracts the sensor's baseline offset to ensure no light maps to true black is __________ __________.

black-level correction

__________ is a repeatable spatial pattern caused by pixel-to-pixel response differences and is reduced using dark or flat-field references.

Fixed-pattern noise

The __________ of a pixel $p = (x, y)$ includes the axis-aligned neighbors: $(x+1, y)$, $(x-1, y)$, $(x, y+1)$, and $(x, y-1)$.

$$N_4(p)$$

Pixels that touch only at a corner are disconnected under $N_4$ connectivity but are considered connected under __________ connectivity.

$$N_8$$

Quantization is an __________ operation because multiple input intensities merge into a smaller number of output levels, meaning the original cannot be uniquely recovered.

irreversible