Reviewing, selecting and evaluating features in distinguishing fine changes of global texture

Abstract

The evaluation of appearance parameters is critical for quality assurance purposes when determining lifetime and/or beauty of textile products. Practical evaluations of appearance are often performed by human visual inspection, which is repetitive, exhausting, unreliable and costly. Thus, computerized automatic visual inspection has been used to alleviate those problems. Several papers have proposed objective mechanisms for quality inspection mostly using texture analysis approaches which are often not robust enough. One of the main issues for robustness of texture analysis approaches is the capability of distinguishing between similar textures. In this paper, we review, select and evaluate texture analysis approaches for distinguishing fine changes of global texture in degradation of textile floor coverings. As a result, we found that the power spectrum, local binary patterns, the texture spectrum, Gaussian Markov random fields, autoregressive models and the pseudo-Wigner distribution provide good descriptors for measuring fine changes of global texture. That is, those features can be used as starting point in applications involving fine changes of global texture, as well as a basis for the development of new methods

Feature selection for surface defect classification of extruded aluminum profiles

“…Such methods for fabric defect detection were also reviewed by Shanbhag et al [22]. Ortiz-Jaramillo et al evaluated different texture classification methods, commonly used by researchers in the field, and provided results with good evidence that power spectrum, local binary patterns, the texture spectrum, Gaussian Markov random fields, auto-regressive models and the pseudo-Wigner distribution are good measures of fine changes in global texture [14].…”