Design And Implementation On Image Denoising For Division Of Focal Plane Polarization Images

Since division of focal plane polarization(DoFP)image sensor is composed of interlaced micro-polarizers with differently oriented polarization axes(i.e.0o,90o,45o,135o),it has its advantages such as compact structure,low manufacturing costs,high imaging speed and carrying out all the polarized information within one photo frame.Interpolation algorithms is essential for the DoFP-based polarization images to reconstruct polarization information.To date,all the developed interpolation algorithms are based on the noiseless polarization model.Owing to the noise is really exist in the image that captured by the DoFP sensor,an image denoising process is needed before interpolation.In this paper,we present a sparse representation based denoising method tailored for DoFP polarization images.The main algorithm steps include: 1)adopt orthogonal matching pursuit(OMP)algorithm for the DoFP polarization images to obtain the sparse representation coefficients;2)update the dictionary atoms via K times singular value decomposition(K-SVD).As random noise cannot be sparsification,it could be separated from the DoFP photo by decomposing the DoFP image into the optimal sparse-representation of the dictionary atoms.Extensive experimental results on various test images show that the PSNR value of the proposed algorithm is ~3dB higher than the state-of-the-art principal component analysis(PCA)algorithm.Additionally,it yields the best visual image quality,which show excellent agreement with the PSNR results.