SURVEILLANCE VIDEO IMPROVISATION BY COLOURING AND ENHANCEMENT OF B/W VIDEO

Abstract

Noise removal, resolution, contrast improvement, and colourization are all used in the colourization of CCTV footage to restore the black and white film medium to its original state. In addition, most CCTV footage is recorded in black and white or with low-quality colours, necessitating colourization for security reasons. We present a framework for dealing with the colourization process semi-interactively in this paper. Our research is centred on attention mechanisms and temporal CNNs (convolutional neural networks). Our suggested source-reference attention allows the model to colourize long movies with an arbitrary number of reference colour images while maintaining temporal consistency without the need for segmentation. Quantitative study demonstrates that our framework outperforms present techniques and that, in assessment to present techniques, the overall performance of our framework is advanced to that of existing approaches.