Abstract: An optical detection technique was applied to on-line detection of surface defects for continuous casting slabs with high temperature. Green linear lasers with high luminance were projected on the slab surface. According to laser monochromaticity, a narrowband color filter was added to lens to prevent slab radiation coming into CCD cameras, and the image contrast was improved. Non-sampling wavelet decomposition was applied to surface images of the slabs, and the scale co-occurrence matrixes of low-pass components and original images were calculated as textural features of images, as well as the scale coccurrence matrixes of high-pass components. A classifier based on AdsBoosting algorithms was developed for classification of defects with textural features. The technique was tested with samples of surface cracks, water marks, slag marks, scales and vibration marks, and the results show that the classification rate of surface cracks is 86.75% and the classification rate of all the samples is 87.16%.