Research On Surface Defect Detection For Smooth Objects With Patterned Surfaces

The objects with smooth surfaces often have good reflective properties, such as steel plate, electronic screen, various printed paper with plastic surface etc. These objects are easy to be scratched due to theirs glossy surface and it is difficult to inspect all the surface defects with a single detection method, especially when the objects' background is complex.The smooth smart cards which maybe contain various surface defects are considered in this paper. The defects are firstly divided into printed type and scratched type. Then different lights and cameras are combined to acquire the images. The printed type defect which is called as A, includes chromatic aberration, spots and deviations etc. While the scratched type defects including scratches, bubbles that generally occur in the card production is called B type for short. A type images are acquired by color camera combined with bowlshaped light source, while B type images are acquired by monochrome camera with parallel coaxial light source. B type defects cannot be observed with naked eyes unless the card is slanted to a certain angle. The defect images can be acquired clearly and completely by this classification method, which is suitable for computer processing algorithm.In the part of computer algorithm, an improved image difference method is proposed to detect the defects for edge and non-edge of A type images respectively, which reduces the influence of the image matching precision on image difference method. So the false defection rate can be decreased effectively. By introducing weighting parameters of between-class variance, an improved Otsu algorithm is proposed to calculate the threshold for segmentation of various bimodal histogram images. Furthermore, an improved method for faint scratch detection under noise interference based on Curvelet transformation is also presented in the paper, which combines the edge enhancement with the edge detection. Then an adaptive Canny operator is introduced to detect the scratch defection. Experimental results show that the proposed algorithm has a better performance on detecting the faint scratch on noisy images.Finally, a real-time defect detection system for smooth cards based on VS2005 platform is introduced, which can meet the requirements of the on-line production defect detection.