TEXEMS: Texture Exemplars for Defect Detection on Random Textured Surfaces

Abstract

We present an approach to detecting and localizing defects in random color textures which requires only a few defect free samples for unsupervised training. It is assumed that each image is generated by a superposition of various-size image patches with added variations at each pixel position. These image patches and their corresponding variances are referred to here as textural exemplars or texems. Mixture models are applied to obtain the texems using multiscale analysis to reduce the computational costs. Novelty detection on color texture surfaces is performed by examining the same-source similarity based on the data likelihood in multiscale, followed by logical processes to combine the defect candidates to localize defects. The proposed method is compared against a Gabor filter bank-based novelty detection method. Also, we compare different texem generalization schemes for defect detection in terms of accuracy and efficiency.

HMM-Based Defect Localization in Wire Ropes – A New Approach to Unusual Subsequence Recognition

“…It is hard to obtain defective examples from a real ropeway, especially due to the strict rules for a regular visual examination [6]. In order to cope with the missing abnormal training material, one-class classification approaches, also known as novelty detection or unusual event detection, have arisen the past years [7,1,3,4,8].…”

“…Xie and Mirmehdi [11,3] employ a Gaussian mixture model (GMM) to detect abnormal variations from the random texture of ceramic tiles. In [12] an OCC approach using a GMM for automatic defect detection in wire ropes is presented.…”

“…Visual inspection of material surfaces has become an important application of pattern recognition [1][2][3][4]. Especially in scenarios, in which a manual inspection implies a high risk for the human life an automatic inspection is highly appreciated.…”

Random Texture Defect Detection Using 1-D Hidden Markov Models Based on Local Binary Patterns

“…Fig. 1 Different patterns of a sound random textured ceramic tile [3]. HMMs that each one models any possible combination of the observation (gray level) sequences (local patterns) extracted from the local windows inside the texture image and each local window has a different texture unit.…”

“…It has shown that, various sizes of LBP codes are necessary to capture sufficient image properties [12]. One way to exploit the multiscale analysis in our methodology is performing the proposed defect detection approach at each scale (level) separately and then combining the detection results from individual scales using an inter-scale post fusion strategy [3]. By doing so, we can use a fixed neighborhood size for all LBPs at each level, and each scale can have its own independent HMMs which the number of them can be different from the other scales.…”