Ernests Lavrinovics

Ernests Lavrinovics

PhD Fellow in NLP at Aalborg University Copenhagen from September 2024. Primary research interests within LLM factuality, multilinguality, knowledge graphs and parameter-efficient fine tuning.

Implementation of Algorithm for Image Feature Extraction: Histogram of Oriented Gradients

Download the Full Paper

Relevant GitHub Repos:

Abstract

Image feature extraction is a vital pre-processing step for image processing tasks such as tracking, object detection or recognition, and image stitching. There are multiple methods for feature extraction such as ORB, SIFT, HOG and others. This contribution describes a from-scratch implementation of Histogram of Oriented Gradients algorithm that is quantitatively evaluated for performance and functionality. Quantitative tests suggest that the implementation suffers from bugs as the gener- ated feature vector has statistically significant differences when compared to production-grade implementation.