Defect Detection in Random Color Textures using the Modified MIA T2 Defect Maps

In this paper we present a new unsupervised approach for the detection of defects in random color textures. This approach is based on the use of the 2 T statistic and it is derived from the MIA strategy (Multivariate Image Analysis) developed in recent years in the field of applied statistics. PCA analysis is used to extract a reference eigenspace from a matrix built by color-textural features of partially overlapped windows or patches inside the input RGB image. For each window of size L L ´ the mean and the variance of each chromatic channel extracted as color features. Also, a compressed version of LBP histograms is used to extract the textural information of each patch. These extracted features make the columns of a data matrix. The same task is performed for each new testing image and the obtained data matrix is projected onto the reference eigenspace obtaining a score matrix used to compute the 2 T images. These obtained images are then converted into defect maps which allow localizing of defective pixels. We present some results from a database of images of artificial stone plates and ceramic tiles.