A Comprehensive Review on Deep Learning Architecture for Pre-processing of Underwater Images

Underwater Image Processing (UIP) is gaining interest in the research community as it is an effective tool for exploring the underwater environment. UIP is used extensively in various fields like topography, mining, archaeology, AUV, aquaculture, structural defects, pipeline maintenance, structural defects, and inspection. On the other hand, processing underwater images is a tedious task as it involves challenges like absorption, color distortion, scattering, attenuation, illumination, blurriness, and turbidity. The pipeline of UIP includes modules viz., underwater image collection, preprocessing, feature extraction, and classification. Among these modules, Preprocessing is a crucial step because its outcome has effects on the other modules. The most commonly used preprocessing techniques are Image enhancement and Image restoration. Most of the traditional underwater pre-processing methods rely on the Image Formation model. Recently, Deep learning methods have been employed to increase the quality of underwater images. This article focuses on the preprocessing techniques for Underwater Images (UWI) proposed in the literature. We first focus on the image formation model for atmospheric and underwater conditions, and then the existing deep learning architectures. We finally summarize the various performance metrics and the benchmark dataset in detail for underwater image pre-processing. The outcome of this review article helps the researchers to understand existing architecture and provides an idea to develop a new framework for pre-processing underwater images.