Regularity Analysis for Patterned Texture Inspection

Abstract

This paper considers regularity analysis for patterned texture material inspection. Patterned texture-like fabric is built on a repetitive unit of a pattern. Regularity is one of the most important features in many textures. In this paper, a new patterned texture inspection approach called the regular bands (RB) method is described. First, the properties of textures and the meaning of regularity measurements are presented. Next, traditional regularity analysis for patterned textures is introduced. Many traditional approaches such as co-occurrence matrices, autocorrelation, traditional image subtraction and hash function are based on the concept of periodicity. These approaches have been applied for image retrieval, image synthesis, and defect detection of patterned textures. In this paper, a new measure of periodicity for patterned textures is described. The Regular Bands method is based on the idea of periodicity. A detailed description of the RB method with definitions, procedures, and explanations is given. There is also a detailed evaluation using the Regular Bands of some patterned textures. Three kinds of patterned fabric samples are used in the evaluation and a high detection success rate is achieved. Finally, there is a discussion of the method and some conclusions. Note to Practitioners-This paper is motivated by the study of a regularity feature for finding common properties in patterned textures. In general, regularity analysis of patterned textures involves two issues: the spatial relationship between intensity values and the repeat distance of a repetitive unit. These issues can also be defined as the periodicity of a patterned texture. However, the traditional periodicity is not effective for developing a patterned texture inspection algorithm. In this paper, a new measure for the regularity of patterned textures is designed for defect detection. It is based on the idea of applying the periodicity as a new principle for patterned texture inspection. A break in periodicity is considered to be a defect in patterned texture inspection. This concept has been applied to the development of a new method called the RB method. The regular band is defined by a moving average and standard deviation of the pixel intensity. It is specialized for defects which have differential intensity changes compared with the pattern on a repetitive unit of patterned texture. The RB method has been found very effective for defect detection of patterned fabric. In a comprehensive evaluation, the detection success rate of the RB approach has reached 99.4% in a total of 166 defective and defect-free images taken from three patterned fabrics. In this paper, the techniques and detection results of the RB method as well as comparisons with other methods are given. The computational time for processing an image of size 256 256 is only 140 ms using the C programming language. This new approach for automated patterned texture inspection is believed to be useful for quality control. It will also make contributions not only to practitio...

Automatic Detection of Texture-defects using Texture-periodicity and Jensen-Shannon Divergence

“…Methods that are found in literature for the inspection of patterned texture images include the traditional image subtraction methods [10][11][12], the method of golden image subtraction (GIS) [5], the method of wavelet-preprocessed golden image subtraction (WGIS) [5], the method of Direct-Thresholding (DT) based on wavelet transform [5], the Bollinger Bands method [6], the Regular Bands method [3], the Local Binary Pattern (LBP) method [7], and the motif-based methods [4,8,9].…”

“…This was done with the help of the moving average and standard deviation [3]. Thresholds obtained from several histogram-equalized defect-free images during the training stage were used for detecting defects on histogram-equalized defective images during the testing stage.…”

“…Inspection of patterned textures belonging to wallpaper groups other than the p1 group is more complicated than that in textures belonging to the p1 group due to a complexity in the design, the existence of numerous categories of patterns, and a similarity between the defect and background [3]. As a result, most of the methods in literature depend on the training stage by using numerous defect-free samples for obtaining decision-boundaries or thresholds prior to the detection of defects (i.e., [3][4][5][6][7][8][9]).…”

“…As a result, most of the methods in literature depend on the training stage by using numerous defect-free samples for obtaining decision-boundaries or thresholds prior to the detection of defects (i.e., [3][4][5][6][7][8][9]). Moreover, methods involving pattern matching and the image subtraction approach perform a pixel-by-pixel comparison, which is inherently sensitive to noises, misalignments, and distortions [8].…”

Homogeneity Cues for Texel Size Estimation of Periodic and Near-Periodic Textures

“…Grigorescu and Petkov [3], determine texel size of periodic and near-periodic images based on the calculation of Renyi's generalized entropies by assuming a square texel for all textures. Other studies for texture periodicity detection are found in the literature, we can mention those based on the wavelet transform [8], autocorrelation [10,9], or regular bands [15].…”