Effective restoration of underwater images to improve local feature matching

Underwater images are characterized by poor contrast, foggy appearance and color shifting and as a result due to these distortions many underwater applications perform poorly. Since many computer vision applications (e.g. SLAM) still rely on effective matching of local features, in this work we analyze this problem for underwater images. We first describe an effective underwater restoration approach that is built on the well-know DCP (dark channel prior). We first perform a simple but effective color correction and then assuming the simplified optical model we estimate the transmission and backscattering light using dark channel prior. In our validation we perform a quantitative evaluation in the context of local feature matching based on the ORB operator. The extensive experiments demonstrate that our approach yields competitive results.