Pattern recognition using inverse resonance filtration

TL;DR

This paper introduces an inverse resonance filtration approach for texture pattern recognition using eigen harmonic decomposition, which is invariant to linear shifts and effective for detecting foreign objects.

Contribution

The paper presents a novel IRF-based method utilizing eigen harmonic decomposition for texture recognition, including two new techniques for estimating 2D EHD parameters amidst signal breaks.

Findings

EHD is invariant to linear shifts in textured images.

Two methods effectively estimate 2D EHD parameters with signal breaks.

The approach successfully detects foreign objects in textured images.

Abstract

An approach to textures pattern recognition based on inverse resonance filtration (IRF) is considered. A set of principal resonance harmonics of textured image signal fluctuations eigen harmonic decomposition (EHD) is used for the IRF design. It was shown that EHD is invariant to textured image linear shift. The recognition of texture is made by transfer of its signal into unstructured signal which simple statistical parameters can be used for texture pattern recognition. Anomalous variations of this signal point on foreign objects. Two methods of 2D EHD parameters estimation are considered with the account of texture signal breaks presence. The first method is based on the linear symmetry model that is not sensitive to signal phase jumps. The condition of characteristic polynomial symmetry provides the model stationarity and periodicity. Second method is based on the eigenvalues problem of matrices pencil projection into principal vectors space of singular values decomposition (SVD) of 2D correlation matrix. Two methods of classification of retrieval from textured image foreign objects are offered.