Research On Image Stitching Under Different Illumination

The image stitching technology is widely used in virtual reality,remote sensing technology,medicine and other fields.And different illumination of natural images is also an essential factor.Therefore image stitching technology research under different illumination has important academic significance and application value.Image stitching is a technology which merged several scene images with overlapping regions into a high-resolution panoramic image.For ensuring the better image mosaic effect,this thesis study the splicing image processing method in the case of different illumination of images.In this thesis,two image illumination processing methods are proposed for splicing images.The method of pixel-wise orthogonal decomposition extracts the color illumination invariant information of images,removing the interference of illumination while preserving the color information of the images for stitching.The second illumination processing method is color constancy of image based on the deep learning.The use of deep convolutional neural networks structure is to extract the illumination information of images.The networks model is trained to learn the average illumination making the splicing image color constancy.Finally,the results of the two kinds of illumination processing algorithms are compared with the analysis on their advantages.In the image stitching stage,this thesis uses the overlapping region matching verification method to eliminate the mismatched pairs of images.Then adopting the cylindrical projecting transformation register the image for keeping the original geometric information of the image.In this thesis,the APAP image stitching datasets of Portland State University carry on the image stitching experiment.The method of illumination processing with deep learning is completed in the Ubuntu system based on the CAFFE network framework.Experiments show that the method proposed in this thesis can get better effect on image stitching under different illumination conditions.