PEDENet: Image Anomaly Localization via Patch Embedding and Density Estimation

TL;DR

PEDENet is an unsupervised neural network for image anomaly localization that combines patch embedding, density estimation, and location prediction to improve detection accuracy.

Contribution

It introduces a novel combination of patch embedding, GMM-inspired density estimation, and location prediction networks for unsupervised anomaly localization.

Findings

PEDENet outperforms state-of-the-art methods on benchmark datasets.

The patch embedding and density estimation approach enhances anomaly detection accuracy.

The location prediction network improves the localization precision.

Abstract

A neural network targeting at unsupervised image anomaly localization, called the PEDENet, is proposed in this work. PEDENet contains a patch embedding (PE) network, a density estimation (DE) network, and an auxiliary network called the location prediction (LP) network. The PE network takes local image patches as input and performs dimension reduction to get low-dimensional patch embeddings via a deep encoder structure. Being inspired by the Gaussian Mixture Model (GMM), the DE network takes those patch embeddings and then predicts the cluster membership of an embedded patch. The sum of membership probabilities is used as a loss term to guide the learning process. The LP network is a Multi-layer Perception (MLP), which takes embeddings from two neighboring patches as input and predicts their relative location. The performance of the proposed PEDENet is evaluated extensively and benchmarked with that of state-of-the-art methods.