Digital Image Processing and Machine Learning Fundamentals

Vocabulary flashcards covering image basics, color and cameras, image processing basics, edges and sharpening, image formats, and machine learning based on the lecture transcript.

Pixel

The smallest individual unit of a digital image.

Image Resolution

The number of pixels in an image.

PPI

Measures pixel density; the number of pixels per inch.

DPI

Measures dot density; the number of printed dots per inch.

Bit Depth

The number of bits used to represent pixel information.

Quantization

Converting continuous values into a limited number of digital levels.

Sampling

Measuring a continuous signal at regular intervals.

Sampling Pitch

The physical distance between neighboring sensor pixels.

Aliasing

Distortion caused by sampling a signal too slowly.

Nyquist Criterion

A signal must be sampled at least twice its highest frequency.

RGB

A color model using red, green, and blue channels.

Indexed Color

Pixels store indices that point to colors in a shared palette.

Demosaicing

Reconstructing full-color pixels from a camera color filter array.

Vignetting

Darkening of an image toward its corners.

Barrel Distortion

Lens distortion that makes straight lines bulge outward.

Pincushion Distortion

Lens distortion that makes straight lines bend inward.

Focal Length

The distance from a lens's principal plane to its focal point.

Optical Power

A lens's focusing strength, measured in diopters. It equals $1$ divided by focal length in meters.

Histogram

A graph showing how many pixels have each intensity value.

Point Operation

An operation that changes individual pixel values based on their intensities.

Homogeneous Point Operation

The same intensity mapping is applied to every pixel, regardless of location.

Convolution

Applying a filter by multiplying corresponding values and adding the products.

Convolution Kernel

A small matrix of weights used to process an image.

Padding

Adding extra pixels around an image before convolution.

Valid Padding

No extra pixels are added; the output image is smaller.

Same Padding

Padding is added to preserve input dimensions when stride is $1$.

Stride

The number of pixels a filter moves at each step.

Linear Filter

A filter that uses weighted sums of input pixel values.

Nonlinear Filter

A filter that does not satisfy the linearity rules.

Gaussian Filter

A linear smoothing filter that reduces noise and small details.

Median Filter

A nonlinear filter that replaces a pixel with the median value of its neighborhood.

Weighted Median Filter

A median filter where some neighboring pixels have more influence because they have higher weights.

Laplacian Filter

A second-derivative filter used to detect rapid intensity changes.

Sobel Filter

A filter used to estimate image gradients and detect edges.

Gradient

A measure of how quickly image intensity changes and in which direction.

Gradient Magnitude

The strength or amount of intensity change.

Gradient Direction

The direction in which intensity changes most rapidly.

Edge Detection

Finding boundaries between different regions in an image.

Zero Crossing

A location where a filter response changes sign, often used for Laplacian edge detection.

Non-Maxima Suppression

Keeps local maximum gradient values and removes weaker neighboring values to make edges thinner.

Image Sharpening

Enhancing edges and fine details in an image.

Unsharp Masking

Blurring an image, subtracting the blur from the original, and adding the resulting mask back to sharpen it.

Normalized Correlation

A method that measures similarity between an image region and a template.

Template Matching

Finding where a smaller image or pattern appears inside a larger image.

Separable Filter

A filter that can be performed as two smaller filters, usually horizontal and vertical.

Sigma ($\sigma$)

Controls the spread of a Gaussian filter; larger sigma means more smoothing.

Lossless Compression

Reduces file size without losing original image information.

Lossy Compression

Reduces file size by removing some image information.

GIF

A lossless indexed-color format that supports up to $256$ colors per frame.

JPEG

A lossy image format commonly used for photographs.

Regression

Predicting a continuous numerical value.

Classification

Predicting which category or class something belongs to.

Multilayer Perceptron (MLP)

A neural network made of layers of connected neurons.

Activation Function

A function that adds nonlinearity to a neural network.

ReLU

An activation function that outputs zero for negative inputs and keeps positive inputs unchanged.

Sigmoid

An activation function that maps values between $0$ and $1$.

Tanh

An activation function that maps values between $-1$ and $1$.