Convolutional Neural Network based Classification for automatic segregation of distorted digital images
Nikitha Saurabh, Prathyakshini, Preethi Salian K · 2022 Second International Conference on Computer Science, Engineering and Applications (ICCSEA) · 2022
In this digital era, passion for photography has seen a tremendous rise owing to the desire to upload photos on social media with the intention of increasing likes or followers. Pre-event and post event shoots for various occasions is a new trend which has increased the demand for professional photographers. With the advent of digital and mobile cameras, a massive number of digital photos can be produced which is only limited by the camera or mobile storage capacity. All the images produced by these cameras necessarily need not be of good quality. Some of the photos may include excessive amounts of blurring which in turn degrades the photo clarity. Manual separation of such low-quality photos can be a difficult and time-consuming task for both professional as well as amateur photographers. In such a situation, automatic detection and classification of blurred images and moving them to a separate folder would save the user from the strenuous job manual separation or deletion of blurred images. This paper proposes a convolution neural network (CNN) based classifier model to identify and classify digital photos based on their clarity. The proposed CNN model was trained on a dataset of 20k digital images (both mobile and digital camera photos) collected from multiple sources and annotated with two classes as blurred and unblurred. The proposed CNN model achieved a validation accuracy of 95% and could efficiently identify the blurred images and classify them.