Learn Computer Vision (for Beginners) Part 2

Explore how linear filtering, including convolution and correlation, reduces noise in grayscale images by applying kernels, revealing how box and gaussian filters create different blur effects.

Explore boundary handling techniques: zero, toroidal, clamp, and mirroring that manage edge effects, and nonlinear filtering, including median filtering and morphological operations such as dilation and erosion on grayscale images.

Explore multi-scale image representations using Gaussian and Laplacian pyramids, built through smoothing and downsampling. Apply them to image compression, object detection, and image blending.

Explore subsampling in computer vision, its effect on aliasing and image quality, and how low-pass Gaussian filtering prevents artifacts for reliable multi-resolution image pyramids.

Explore how edge detection identifies boundaries by analyzing intensity changes using first and second derivatives, Gaussian smoothing, and derivative filters, with hysteresis for robust localization.

Explore how the Gaussian function and its derivatives reveal image edges, from zero-order to first and higher orders, highlighting local maxima, minima, and zero crossings.

Extend edge detection from 1D to 2D using partial derivatives in x and y, with Sobel and Prewitt filters applying Gaussian smoothing.

Explore the optimal canny edge detector and its use of Gaussian smoothing, gradient magnitude, non-maximum suppression, and hysteresis thresholds to balance detection and localization.

Explore how non-maximum suppression and hysteresis refine edge detection by linking strong and weak edges, while analyzing the impact of sigma and SNR on Canny edge detection.

Learn how the laplacian of gaussian detects edges by using the Mexican hat shape and the second derivative of gaussian, and see how DoG approximates LoG with different sigma scales.

Learn view-based recognition using template matching to compare features across images with cosine similarity and normalized correlation, while handling rotation, scaling, and intensity variations.

Learn how subspace models reduce high-dimensional data to a lower-dimensional subspace via linear transformations and PCA, enabling dimensionality reduction by identifying basis vectors to balance accuracy and complexity.

Explore principal component analysis by visualizing how a cluster centered at the origin reveals maximum variance along a principal direction, links to linear dimensionality reduction, and error minimization.

Explore principal component analysis by deriving projection variance with a covariance matrix, eigenvalues, and least squares estimation, then select the subspace dimension from eigenvalues for PCA applications like eigenfaces.