A two-layer Conditional Random Field for the classification of partially occluded objects

TL;DR

This paper introduces a novel two-layer Conditional Random Field framework designed to improve the classification of partially occluded objects in images, demonstrating effectiveness on aerial and street-view datasets.

Contribution

The paper presents a new two-layer CRF model specifically tailored for classifying partially occluded objects, enhancing image labeling accuracy in complex scenes.

Findings

Effective in classifying occluded objects in aerial images

Outperforms existing methods on urban street-view datasets

Demonstrates robustness in diverse imaging conditions

Abstract

Conditional Random Fields (CRF) are among the most popular techniques for image labelling because of their flexibility in modelling dependencies between the labels and the image features. This paper proposes a novel CRF-framework for image labeling problems which is capable to classify partially occluded objects. Our approach is evaluated on aerial near-vertical images as well as on urban street-view images and compared with another methods.