Stop-band Energy Constraint for Orthogonal Tunable Wavelet Units in Convolutional Neural Networks for Computer Vision problems

TL;DR

This paper proposes a stop-band energy constraint for orthogonal tunable wavelet filters in CNNs, improving image classification and anomaly detection accuracy, especially on texture-rich datasets.

Contribution

It introduces a novel stop-band energy constraint for wavelet filters integrated into CNNs, enhancing their performance on various vision tasks.

Findings

Achieves 2.48% accuracy gain on CIFAR-10

Improves texture dataset performance by 13.56%

Outperforms existing methods in anomaly detection

Abstract

This work introduces a stop-band energy constraint for filters in orthogonal tunable wavelet units with a lattice structure, aimed at improving image classification and anomaly detection in CNNs, especially on texture-rich datasets. Integrated into ResNet-18, the method enhances convolution, pooling, and downsampling operations, yielding accuracy gains of 2.48% on CIFAR-10 and 13.56% on the Describable Textures dataset. Similar improvements are observed in ResNet-34. On the MVTec hazelnut anomaly detection task, the proposed method achieves competitive results in both segmentation and detection, outperforming existing approaches.