Anti-noise light field depth measurement using specular highlight removal

•The scene prior and the modified Poisson mixing method correct the diffuse reflection component.•The threshold method classifies specular pixel candidates to improve the operating efficiency.•The noise-aware optimization strategy removes the scene noise of the light field. Due to their ability to simultaneously record spatial and angular information about the scene within a single shot, light field cameras have a distinct edge in depth measurement. Both specular highlight and noise make it difficult to measure the depth of the light field. To address this problem, we propose an alpha hybrid method based on the dichromatic reflection model (DRM) and scene prior to remove specular reflection components and optimizing the cost using a noise-aware strategy. The quantitative results demonstrate that compared with the best performance Shen, our method can reduce the MSE by 40.68 %, increase the PSNR and SSIM by 9.65 % and 28.18 %, respectively, in the highlight challenging noisy scenes of HCI datasets. The qualitative results show that we can restore the highlight and preserve various fine structures of the real light field. Therefore, our method has significant advantages in high-noise scenes and can better deal with depth measurement in non-Lambert scenes.