Fabric defect segmentation using multichannel blob detectors
In this work, a multi-channel filtering technique based on the real Gabor functions for segmentation of local texture defects was presented. The technique has been developed and evaluated for on-line detection of local defects in textile webs. One of the advantages of multi-channel filtering approach over other textural feature extraction approaches that use a small window size is its ability to segment both fine and coarse texture defects. This is accomplished by segmenting fine and coarse texture defects on different scales (multi-channel). The proposed approach was motivated by earlier work on mechanisms in visual cortex of mammals. The algorithm uses real Gabor functions instead of complex Gabor functions. Image fusion technique based on Bernoulli’s rule of combination was employed to integrate information from different channels. This approach offers high detection rate and low false alarm. More details on this work can be found in the paper from Optical Engineering.
Figure 1: Bank of sixteen real Gabor filters employed to sample the textured features.
Figure 2: Supervised defect-detection using Gabor filter.
Figure 3: The effect of sensitivity control (Low, Medium, and High) on segmentation of defects.
