Underwater Image Enhancement on Low-Cost Hardware Platform

Abstract

In the recent years several cost effective and versatile remote operated vehicles (ROV) have been developed. However, these drones are equipped with low powerful hardware platforms. In this work we present an underwater image enhancement solution that works effectively on such hardware platforms. The approach is built on the underwater optical model and estimates locally the backscattered light component. In order to compute an optimal image patch size, we estimate two complementary values of the local backscattered light (an estimate for a large patch size and one for a small patch size) that are averaged in an optimal value. The transmission map is computed based on the well-known dark-channel prior (DCP) [1]. Finally, the results are yielded by inverting the simplified optical model using the estimated values of the local backscattered light an the trans-mission. The method was implemented and tested on Raspberry Pi. Our extensive experiments show that the proposed technique is computationally effective but also competitive compared to several specialized techniques.