Gaussian Pyramid and DOG Generation in SIFT Operator

The creation of Gaussian pyramid and Difference of Gaussian (DOG) constitutes the foundational stage of the SIFT operator. In this phase, we first construct a Gaussian pyramid, which consists of multiple image layers created by applying Gaussian blur filters with progressively increasing standard deviations (sigma values). Each octave in the pyramid contains images at different scales, typically implemented through iterative convolution operations with Gaussian kernels. Subsequently, we compute the differences between adjacent layers within the Gaussian pyramid to generate the DOG pyramid. This difference operation effectively highlights potential keypoint locations by approximating the scale-normalized Laplacian of Gaussian. The computational process involves subtracting corresponding pixel values between consecutive Gaussian-blurred images at each octave level. This initial stage serves as the starting point for the SIFT algorithm, providing the multi-scale representation necessary for subsequent feature detection, descriptor extraction, and matching operations. The DOG pyramid particularly enables efficient identification of scale-invariant interest points through local extrema detection across scale space.