Research On The Model And Solution Algorithm Of Color Image Demosaicing

It is common that the single sensor CCD(electronic coupling element)is used as image sensor in digital camera.While taking photographs,each pixel only gets one color,and the image is made up of red,blue and green three colors,to get the full color image,we must restore the other two missing components,this process is called to demosaicking.The quality of restoration image plays an important role for image subsequent processing in computer vision system such as image recognition,image segmentation and so on,so the effective demosaicking algorithms has important application value and realistic significance.We will study on the color image demosaicking according to the correlation between the three color channels in view of the existing methods there are false color,zipper effect in the restoration image.In this paper,the main works includes:(1)We introduce some knowledge about CFA(Color Filter Array)image demosaicking,and review some existing demosaicking algorithms.(2)We set up the model based on total variation regularization and the correlation between the three color channels,then making the fidelity term and the regularization term in the RGB space and the brightness chromaticity space respectively,turning demosaicking problems into image reconstruction problems,and solving it with efficient solution algorithm,a primal-dual fixed point algorithm for convex separable minimization.Finally,we use ARI algorithm to improve the solution.We test the new algorithms through two images of the standard test set: Kodak Photo CD and Mc Master image,and compare with the current six outstanding results.Our experiments show that our algorithms are better on these two image sets.(3)Established the demosaicking sparse model based on multi-scale tight wavelet frame in luminance chrominance space and RGB space,and to see which space is good for demosaicking.Kodak data set are relatively smooth,and its saturation is lower compared with Mc Master image set,while Mc Master images have more features and sharp edges.Based on the mean saturation of an initial guess,our proposed method can automatically choose the better approach out of analysis approach or synthesis approach.Finally we use the current optimization algorithm,a primal-dual fixed point algorithm for convex separable minimization,to solve the these models.Through numerical experiments,we can see that demosaicking with RGB space is better than luminance chrominance space while using tight wavelet frame.