MATLAB Implementation of Gaussian Pyramid with Algorithm Explanation

The Gaussian pyramid implementation in MATLAB utilizes the function Gscale(img, levels, gsize, sigma), which accepts four key parameters for multi-scale image processing. The input parameter 'img' represents the source image to be processed through the pyramid structure. The 'levels' parameter defines the total number of pyramid layers to generate, while 'gsize' specifies the dimensions of the Gaussian filter kernel (typically an odd-numbered tuple like [5,5] or [7,7]). The 'sigma' parameter controls the standard deviation of the Gaussian distribution, determining the degree of smoothing applied at each level. This algorithm constructs a multi-scale image representation by iteratively applying Gaussian smoothing and downsampling operations. At each pyramid level, the implementation performs convolution using a 2D Gaussian kernel generated with MATLAB's fspecial('gaussian', gsize, sigma) function, followed by subsampling through pixel decimation (commonly by a factor of 2). The process creates a hierarchical image set where successive levels represent progressively coarser resolutions of the original image. Key implementation steps include: 1. Initializing the pyramid with the original image as level 0 2. For each subsequent level, applying Gaussian filtering using conv2() with symmetric padding 3. Downsampling filtered images using imresize() with 'nearest' or 'bilinear' interpolation 4. Storing results in a cell array or structure for multi-scale access The resulting Gaussian pyramid serves fundamental computer vision applications including scale-invariant feature detection (SIFT), image blending in Laplacian pyramids, and multi-resolution analysis for object recognition. The Gscale function encapsulates these operations while providing control over pyramid depth and smoothing characteristics through customizable parameters.