Local Walsh-Hadamard spectra in video sequence image classifiers

Abstract

A method and software have been developed for classifying video images. The Walsh-Hadamard transform was used to form descriptors of “weak” classifiers. The software allows you to create a database of class features, determine the two-dimensional Walsh-Hadamard spectrum of segments, train fully connected neural networks, and perform exploratory analysis to study the relevance of two-dimensional spectral coefficients. Application of the method and software was carried out on ultrasound images of the pancreas.