IMPROVING THE IMAGE RECOGNITION PERFORMANCE BY SELECTING INFORMATIVE REGIONS
OHCHUL SHIN *
Computer Sciences Division, STEM Science Center, 111 Charlotte Place Suite#100, Englewood Cliffs, NJ 07632, US State.
*Author to whom correspondence should be addressed.
Abstract
The Convolutional Neural Network (CNN)s are machine learning algorithms that mimic the activity of an actual brain to convey some computer-generated procedure. One of the most frequent applications of these CNNs is image recognition software. In this study, an ANN was first generated using the number of images and then tested through different confusion matrices to determine the accuracy of said ANN. After evaluating the accuracy, a collection of skin lesion images ran through the Neural Network with masking applied to these images to determine if the ANN was accurate in classification. The study demonstrated that mask four was the most accurate at 35% through the testing trials.
Keywords: Classification, confusion matrix, machine learning, neural networks, skin lesions