Research On Image Super - Resolution Reconstruction Algorithm In Intelligent Traffic

In the field of computer vision, the image super-resolution restoration is the key step to the achievement of the functions such as object identification, target tracking, image segmentation, image fusion and so on with high efficiency. MAP (maximum a posterior) super-resolution restoration algorithm is widely used because of its easiness of the introduce of prior model, the great ability of denoising, and the uniqueness of solution, and it can restore the detail information greatly. While, the image restored by traditional MAP image super-resolution restoration algorithm will be smooth and the feature of margin will lose. And it has great significance to the restoration of complex video image and the following object identification, target tracking, image segmentation. And traditional MAP image super-resolution restoration algorithm can only restore the image under good conditions, those taken under severe environment will have a bad result by using it.Aiming to the problems before, an improved MAP image super-resolution algorithm has been proposed in this paper. This algorithm has improved the image matching and image interpolation during image super-resolution restoration procedure. The concrete procedure is described as bellow:(1)This paper mainly study on the theory of image super-resolution restoration based on the frame of MAP algorithm. ORB(Oriented FAST and Rotated BRIEF) matching algorithm has been applied in the procedure of movement registration to achieve the feature matching of image. The MSR(Multiple-Scale Retinex) algorithm has been applied to the image enhancement to "defog" the image because the feature matching will be inaccurate under the condition of greasy weather and the invisible illumination. The theory of fuzzy has been applied to correct the color distortions during the image enhancement according to different membership. The initial image, the image after MSR and the image after fuzzy MSR have been matched by ORB algorithm. The result indicates the high accuracy, the stable property and the applicability of the occasion with complex scene of the improved algorithm.(2)The traditional image interpolation algorithm is based on gray image which will lead to the missing of color information of image, this paper has proposed an interpolation algorithm based on color image. The image should be divided into R,G,B images before interpolation and been interpolated respectively to solve the interpolation problem of color image.(3)This paper has improve the link of image interpolation aiming to the problem that the image will has a zigzag and blurry margin interpolated by traditional MAP restoration algorithm by using Curvelet interpolation algorithm. For a higher accuracy, the margin of image is detected by color Canny based on multi-gradient algorithm, and the image is divided into margin area and smooth area and these two areas are interpolated by different algorithm. Within the smooth area of the image, the direction factors are applied in the interpolation because they have different weight in different directions which can solve the problem that the image has a zigzag and blurry margin combined with bicubic interpolation algorithm. And the margin area of images is interpolated by bilinear interpolation. And then the image has been interpolated by different algorithm and been restored by MAP algorithm. The result indicates that the improved algorithm can keep the color information and the marginal feature better, and it can reduce the appearance of the zigzag and blur of margin.This paper has improved MAP algorithm aiming to the problem that the color image and images taken under severe environment will have a bad result restored by traditional MAP algorithm, which has a good result.