Improving Machine Learning Classification Accuracy in Noisy Environment

Image classification is one of the central research in machine learning that has wide application in many fields. Many factors can adversely impact the classification accuracy, including particularly unexpected noise in images. In this paper, we analyze methods to improve image classification accuracy. Particularly, we analyzes the efficacy of two techniques. First, we find that using noise reduction techniques can enhance image classification accuracy, regardless of whether noise has been added to the original images. Second, we find that weighted voting can be used to improve image classification accuracy even when the voting set is not be entirely independent. Additionally, we identify an inconsistency in classification accuracy and image fidelity, which could potentially be exploited for adversarial attacks.