An Energy-Based Back-Propagation Algorithm for Image Recognition
Ahmad Sharieh, Ahmad Al-Yahia · Dirāsāt. Al-ʿulūm al-asāsiyyaẗ · 2010
It is important to recognize images such as fingerprints, even if it’s with noise, rotated or with missed area. It is difficult to recognize such images using the Standard Back-Propagation algorithm (SBP) neural network with high accuracy identification rate. Thus, this paper presents an Energy based Back-Propagation algorithm (EBP). The energy function is used with the convergence process to extract the nearest image for the unknown tested image. The EBP algorithm combines the fuzzy logic and the back-propagation neural network. It uses the fuzzy logic to convert the digital image into binary image and the back-propagation net for the identification process. The EBP algorithm shows considerably better performance in terms of time of learning, time of convergence and size of input image compared to the SBP algorithm. For the tested samples, the time of learning and convergence was 8.2 seconds, using the EBP, compared to 1312.7 seconds, when using the SBP. The EBP can handle images of sizes up to 600 by 600 pixels, while the SBP can handle images of sizes up to 10 by 10. The EBP showed accuracy of recognition close to 100% for images without noise and it is better than the SBP by 20% to 80% in cases with noise. The performance of the EBP was investigated on images of different noise ratios, rotated images and images with missing areas. It is found that the EBP can recognize images that were effected by noise, rotation and missing parts.