Unsupervised Defect Detection Research Articles

GPU Architecture for Unsupervised Surface Inspection Using Multi-scale Texture Analysis

Surface inspection in manufacturing scenarios is strongly related to accuracy and runtime requirements. To ensure high accuracy and reliable defect detection results, in many applications the environment will be modified with respect to constant illumination and well defined system behavior. In these cases early vision algorithms like edge detection or thresholds are applied for surface inspection which are not complex and runtime intensive. In more complex scenarios with changing illumination conditions more elaborate image processing techniques are needed to ensure reliable defect detection, which leads to more runtime intensive algorithms. To overcome this challenges time-consuming operations can be transferred to additional hardware to satisfy the strong runtime constrains even for complex image processing techniques. The graphics processing unit (GPU) as a co-processor offers great potential and massively parallel computing power to enable real-time application of complex computing steps in production scenarios. We introduce a GPU based implementation of unsupervised defect detection on textured surfaces. Evaluation on an artificial dataset confirms excellent defect detection results and real-time performance.

Statistical approach to unsupervised defect detection and multiscale localization in two-texture images

We present a novel statistical approach to unsupervised de- tection and localization of a chromatic defect in a uniformly textured background. The test images are probabilistically modeled using Gauss- ian mixture models, and consequently defect detection is posed as a model-order selection problem. The statistical model is estimated using a modified Expectation-Maximization algorithm that aids in faster conver- gence of the scheme. A test image is segmented only if a defective region/blob has been declared to be present, and this improves the effi- ciency of the entire scheme. This work places equal emphasis on defect localization; hence, an elaborate statistical multiscale analysis is per- formed to accurately localize the defect in the image. The underlying idea behind the multiscale approach is that segmented structures should be stable across a wide range of scales. The efficacy of the proposed approach is successfully demonstrated on a large dataset of stained fabric images. The overall detection rate of the system is found to be 92% with a specificity of 95%. All of these factors make the proposed approach attractive for implementation in online industrial applications.

Automated defect detection system using wavelet packet frame and Gaussian mixture model

This paper proposes an approach for automated defect detection in homogeneous textiles using texture analysis. The texture features are extracted by the wavelet packet frame decomposition followed by the Karhunen-Loève transform. The texture feature vector for each pixel is used as an input to a Gaussian mixture model that determines whether or not each pixel is defective. The parameters of the Gaussian mixture model are estimated with nondefective textile images in supervised defect detection. An approach for unsupervised defect detection is also presented that can identify the heterogeneous subblocks on the basis of the Kullback-Leibler divergence between two Gaussian mixtures. The proposed method was evaluated on 25 different homogeneous textile image pairs, one of each pair with a defect and the other with no defect, and was compared with existing methods using texture analysis. The experimental results yielded visually good segmentation and an excellent detection rate with a low false alarm rate for both supervised and unsupervised defect detection. This confirms the validity of the proposed approach for automated defect detection and localization.