Tablets Vision Inspection Approach Using Fourier Descriptors and Support Vector Machines

In this paper, an efficient approach for tablets vision inspection is proposed, which can detect missing and broken individual tablets in each blister after they are sealed. The images of tablets in blister can be obtained clearly using multi-lights. From these images the regions of tablets are segmented through thresholding method, and the tablets' shape contours are obtained by Canny edge detector. The Fourier descriptors of closed contours are carried out to extract the tablets' feature and a new classification algorithm based on support vector machine for quality level of tablets packing are presented. The experimental results showed that 88.9% classification accuracy was achieved with a linear kernel, 95.6% with a polynomial kernel, and 99.2% with a Gaussian radial basis function kernel. The machine vision system developed has a large potential to assist in the inspection of tablets quality level classification.