Colourization of Black and White Images using Convolutional Autoencoder

Image colourization, the process of adding colour to grayscale images, remains a significant challenge in computer vision with diverse applications in digital media. This paper presents the development and evaluation of a convolutional autoencoder model designed for automatic image colourization. The methodology involves converting grayscale images into the LAB colour space, where the ‘L’ channel is used as input to predict the ‘A’ and ‘B’ channels. The model architecture integrates convolutional layers for feature extraction and upsampling layers for image reconstruction. The Adam optimizer is employed to minimize the Mean Absolute Error (MAE) during training. Utilizing a landscape image dataset and incorporating data augmentation techniques, the model achieved an accuracy of 85.34%, demonstrating effective colour preservation and detail enhancement. This approach automates the colourization process with high efficiency and provides valuable insights for practical applications in digital media.