Enhancing Underwater Colour Restoration using a Detectron-Autoencoder Hybrid Model

Abstract

Image colorization in underwater environment has become a useful process for submarine operations. Colorized images can enhance the detection of specific targets or anomalies underwater, recently, most researchers of oceanology use high quality underwater images to make detailed analysis of submarine resources. This paper presents deep learning method for underwater images colorization. Studies show that training the deep learning models on a large dataset of underwater images and corresponding ground truth colour images improves objects definition and colorization in the image; hence, our model learns to segment objects from the underwater image, increase the intensity of degraded colours, merge the segmented object with the original image and to another image has different objects in the same context. This paper illustrates image processing techniques such as colour correction and restoration, for imaging challenges and extensive experiments that illustrate the high performance of the proposed model compared to existing state-of-arts, recovering true colours and enhancing overall image quality. The proposed model achieved good results in image colorization with accuracy of 92% that indicates model performance in classifying the targeted object.