Defect Detection in Fabrics Using Local Binary Patterns - Advances in Image and Graphics Technologies - page 279

Fig. 3. Detection algorithm process

The defect image is divided into overlapping detection windows of size W d × W d
pixel as the same method as processing defect-free image. The LBP mask is slided
according to pixel in sequence for each detected window. The feature vector S k of each
detected window can be calculated at the same time. Calculate the similarity of each
detection window and the entire image according to Eq.8. Compare L k with the thre-
shold T . If L k > T , the detected window is judged as the defect window, in which all the
pixels located are set as 255. Otherwise the detected window is judged as the defect free
window, in which all the pixels located are set as 0.

4 The Experimental Results

The experiment is conducted with MATLAB 7.6.0. As experimental subjects to verify
the enforceability of the proposed method, the test samples are selected from the
TILDA database and Henry Y. T. Ngan of the Industrial Automation Research La-
boratory of Hong Kong University. The size of each picture is 256 × 256 pixel.
Process the sample images with the steps of the training phase and the testing phase
described above. LBP 8, 1 , LBP 16, 2 and LBP 24, 3 are respectively used to process the images
to extract their feature values and detect the defects. In addition, to develop the accuracy
of the detection, the overlapping step is 2. In the paper, ten kinds of backgrounds and
defects images are analyzed as examples. The detection results are presented in Fig.4.