Abstract:

Objective Regularity (periodicity), as one of the most important attribute features of texture images, is a key index for classifying and describing texture images. Relating to the unique production process of woven fabrics, the texture of woven fabric images exhibits a high degree of regularity, which is a typical structural texture. Therefore, utilizing regularity to describe the surface attributes of fabric textures is of great practical significance in the fields of fabric classification, fabric retrieval, and fabric apparent quality detection and evaluation.

Method This paper first proposes four metrics for characterizing the regularity of fabric textures based on the spatial arrangement of primitive elements and human visual perception factors (grey scale contrast, structural symmetry and edge consistency). Then, significant peaks and valleys are extracted by threshold division on autocorrelation Pearson coefficients and distance matching function curves, and the relationship between peaks and valleys is used to achieve the characterization of the above metrics. Finally, the weighted summation is used to perform the comprehensive characterization of texture regularity.

Results The overall regularity metrics are validated using different types of fabric samples and some structured textures from the Brodatz Album. The results show that for fabric texture, the proposed metric is able to achieve the characterization of the regularity of fabric texture and distinguish effectively textures with different degrees of regularity. In addition, the comparison with other methods shows that the proposed metric is more in line with the visual perception law of the human eye and has a higher accuracy rate. Under the understanding that the textures in the Brodatz Album usually show a high degree of regularity and have a relatively large and unique size of the periodic unit, two different combinations of the weights are discussed in order to find the most suitable weights, and calculate the overall degree of regularity of the texture with the proposed metric under the weight. It is concluded from the discussion that the proposed metric can also achieve the characterization of the regularity of fabric textures. The results show that the proposed metrics are capable of extracting structural textures other than near-regular fabric textures.

Conclusion The texture regularity index is subdivided from the perspective of the visual perception of the human eye, so as to achieve the characterization of texture regularity in a more comprehensive and detailed way. The experimental results show that the proposed metrics are not only applicable to fabric textures with small and soft sizes, but can also characterize the regularity of various specific types of texture images by combining the importance of different metrics and applying different weights. In the future, the proposed metrics can be considered to be combined with other attribute features of texture images for practical applications such as texture classification and defect detection.

Key words: fabric texture, texture regularity, contrast, autocorrelation Pearson coefficient, human visual feature