A novel gravitational search algorithm for multilevel image segmentation and its application on semiconductor packages vision inspection

Abstract

Multilevel image segmentation is an important technique and indispensable process in vision inspection on semiconductor packages to sort out defective products from the qualified ones, classify and identify the defect types. Conventional multilevel image segmentation methods are computationally expensive, and lack accuracy and stability. To address this issue, this paper proposes a novel gravitational search algorithm (NGSA) for multilevel image segmentation. Two major improvements to the update mechanism (UM) have been made in NGSA, i.e., adaptive gravitational constant and normal mutation of global best agent, to help agents jump out of local optima and improve the calculation accuracy. The experimental results based on multilevel Otsu criterion demonstrate that the proposed NGSA can obtain optimal multilevel thresholds for the quad flat non-lead (QFN) defect images and the segmentation results are promising. Three different methods, firefly algorithm (FA), cuckoo search (CS) and gravitational search algorithm (GSA), are compared with the proposed method. Numerical illustrations show that the proposed NGSA outperforms FA and GSA, and performs as well as, or is better than CS in solution quality, computational efficiency, and operation stability. Hence, NGSA in combination with multilevel Otsu criterion can be accurately and efficiently used in multilevel image segmentation of vision inspection on semiconductor packages.