We present the design, implementation, and evaluation of SeaScan, an energy-efficient camera for 3D imaging of underwater environments. At the core of SeaScan's design is a trinocular lensing system, which employs three ultra-low-power monochromatic image sensors to reconstruct color images. Each of the sensors is equipped with a different filter (red, green, and blue) for color capture. The design introduces multiple innovations to enable reconstructing 3D color images from the captured monochromatic ones. This includes an ML-based cross-color alignment architecture to combine the monochromatic images. It also includes a cross-refractive compensation technique that overcomes the distortion of the wide-angle imaging of the low-power CMOS sensors in underwater environments. We built an end-to-end prototype of SeaScan, including color filter integration, 3D reconstruction, compression, and underwater backscatter communication. Our evaluation in real-world underwater environments demonstrates that SeaScan can capture underwater color images with as little as 23.6 mJ, which represents 37X reduction in energy consumption in comparison to the lowest-energy state-of-the-art underwater imaging system. We also report qualitative and quantitative evaluation of SeaScan's color reconstruction and demonstrate its success in comparison to multiple potential alternative techniques (both geometric and ML-based) in the literature. SeaScan's ability to image underwater environments at such low energy opens up important applications in long-term monitoring for ocean climate change, seafood production, and scientific discovery.