Image Classification Based on Quality of Pixels

Anomaly detection for CCTV Surveillance Systems to give Alerts and Auto-Alerts when no one monitors the CCTV is most important, making it easy to keep track of events. The vital component of training a machine learning model is data. The data here are images from various timestamps and from every camera present on the industry campus. If the data fed is flawed then the model can generate false alarms, making it challenging to identify true anomalies. So, in this paper, there are techniques to either discard the images that cannot be enhanced or enhance them in order to increase the quality of images fed to the anomaly detection model. In addition to this, we have used Kurtosis and Brisque values as parameters of every image and trained the values on different ML and DL models such as Logistic Regression, Decision Tree, Naïve Bayes, VGG16, CNN, AlexNet, and LeNet5.