Research On Image Processing And Algorithm Based On Differential Manifold

Because of the influence of the external environment, the image will suffer some damages of different types and diverse degrees, forming multiple kinds of degraded models. The quality of the video will also decline by the destruction of the image damage caused by environment factors. Finding the efficient way to restore the damage image and video from original files plays an important role in traffic, medical, military and other fields. In this paper, a more efficient algorithm is proposed in color image denoising, deblurring, interpolation, super-resolution reconstruction and dehazing reconstruction, to realize the super-resolution reconstruction of color video.In terms of color image reconstruction, a color image differentiable manifold regularized model based on Nambu functional is proposed and using explicit format to iterative calculation. This model overcomes edge color dithering of the reconstructed color image. It is shown in the experiments that this algorithm has excellent results in image and video reconstruction. However, the explicit format has large calculating quantity and extends the computing time, so further applying the semi-implicit LOD (locally one-mensional) splitting scheme based on Beltrami model to iterative computation. Experiments show that semi-implicit scheme can reconstruct the damaged color image more efficiently.In terms of color video super-resolution reconstruction, genetic algorithm is used to dynamically search parameters and SURE criterion is applied to determine optimal parameters for each frame of video, finally applying the optimal parameters to the corresponding damaged frame by semi-implicit scheme, so as to realize the color video reconstruction process. Simulation results show that the algorithm can excellently reconstruct the video with small difference between adjacent frames.In terms of color image dehazing, the optimized semi-implicit LOD splitting scheme is used to study image dehazing. Experiments demonstrate that the proposed algorithm has higher operation efficiency and can better preserve the edge details and texture of the reconstructed image.