OCR for Unreadable Damaged Characters on PCBs Using GSC Algorithm and kNNClassifier

摘要

In this paper, we propose to change the actual implemented pattern matching method to have optical character recognition by implementing the Gradient, Structural, Concavity (GSC) algorithm to extract the features of damaged, unreadable or incomplete numerical digit characters from images on printed board circuits (PCBs). Grayscale color images are acquired from a charge-coupled device (CCD) camera, assembling a dataset of 500 matrix images samples for the character digits from 0 to 9. The GSC feature extraction method is applied to get the characteristics that will be used in the character recognition step. Experimental results show that applying GSC algorithm to extract the features and using k-Nearest Neighbor (kNN) Classifier with the Euclidian Distance can improve optical character recognition (OCR) detectability of damaged characters from actual 95% to more than 97% in early tests.