IPSO based binarization processing in uneven illumination images for billet defect detection

In the automatic grading system of billet defects based on images, the first task is to segment the defect parts from the billet images. But the uneven illumination due to the reflex on the billet images makes the classical image segment methods, for example, the OTSU binarization ineffective. This paper analyzed the uneven illumination characteristic on the billet images and proposed an illumination preprocessing and image-partition scheme. To get the binarizaiton threshold of each partitioned image, this paper presented and utilized an improved particle swarm optimization (IPSO) algorithm with the inertia weights declining exponentially and randomly. The experiment results of the billet images show that the proposed illumination preprocessing and IPSO based binarization method are effective and efficiency. The effects of image segmentation reach the requirement of the billet defect grading. The proposed image segment scheme rarely need parameters adjusted by human beings, so can guarantee the objectivity and standardization of the automatic grading system.