Automated Detection of Irregularities on Magnetic Audio Tapes Using Frame Differencing

Preserving the integrity and authenticity of magnetic audio tape recordings involves identifying any changes, both intentional and unintentional, that occur in the tape medium. This project aims to develop an automated process within the framework of the Moving Picture, Audio, and Data Coding by Artificial Intelligence (MPAI) Context-based Audio Enhancement (CAE) standard, which includes Audio Recording Preservation (ARP) as a use case. The methodology employs frame differencing techniques to compare consecutive video frames, enabling the detection of subtle alterations. Following this comparison, filtering and image processing operations are applied to highlight potential irregularities within the audio tape recordings. Various processed frames, including direct screenshots, difference frames and thresholded frames are extracted and a dataset is created to represent different types of irregularities. The content of the dataset is inputted into a ResNet-based deep learning module for automated classification of detected irregularities. The proposed method seeks to improve the automation and precision of irregularity detection in magnetic audio tape videos, advancing the field of digital archiving and preservation.