Colourization of Black and White Images using Deep Learning

Abstract

Colorization is the process of transforming grayscale photos into colour images that are aesthetically appealing. The basic objective is to persuade the spectator that the outcome is genuine. The majority of grayscale photographs that need to be colourized are of nature situations. Over the last 20 years, a broad range of colorization methods have been created, ranging from algorithmically simple but time- and energy-consuming procedures due to inescapable human participation to more difficult but also more automated ones. Automatic conversion has evolved into a difficult field that mixes machine learning and deep learning with art. The purpose of this study is to provide an overview and assessment of grayscale picture colorization methods and techniques used on natural photos. The study categorises existing colorization approaches, discusses the ideas underlying them, and highlights their benefits and drawbacks. Deep learning methods are given special consideration. The picture quality and processing time of relevant approaches are compared. Different measures are used to judge the quality of colour images. Because of the complexity of the human visual system, measuring the perceived quality of a colour image is difficult. Multiple metrics used to assess colorization systems provide results by calculating the difference between the expected colour value and the ground truth, which is not always consistent with image plausibility. According to the findings, user-guided neural networks are the most promising category for colorization since they successfully blend human participation with machine learning and neural network automation.