Research On Text Image Mosaic Technology Based On SIFT Algorithm

Image mosaic technology has become one of the more and more popular research hotspots in the field of digital image processing because it can combine multiple images to form a new image.Its application range includes virtual reality technology,artificial intelligence recognition,unmanned aerial vehicle mapping,satellite image fusion,remote sensing observation and other fields.Image mosaicing is to transform two or more images with overlapping area in spatial position,so that multiple images are in the same spatial coordinate system,and then multiple images are mosaiced and fused into one image with wide viewing angle and high resolution.The information of the fused image is similar to that of the multi images before fusion,which not only has a wider field of vision and a higher resolution,but also has most of the image information,without losing the main image information.In image mosaic technology,image registration is the key step.Image registration can be divided into gray-scale and feature-based registration methods.Because the gray-scale image registration method is computationally expensive,time-consuming,and inefficient,and the effect of registration is not as good as the latter,this paper chooses the SIFT feature-based registration method for text image mosaic and optimizes its matching algorithm.The main purpose of image fusion method is to eliminate the stitching line after text image stitching and fusion.According to the character of text distribution of text image,this paper uses regional segmentation to realize the segmented fusion of text image,and the effect is good.As for the text image stitching and fusion algorithm based on the region segmentation,this paper divides the text image region according to the blank line gap of the text,adjusts the brightness value of each region,so that the brightness value between the reference image and the image to be registered tends to be the same,and realizes the text image fusion by using the weighted average fusion algorithm.Based on the existing feature splicing method based on SIFT(Scale Invariant Feature Transformation)algorithm,this paper optimizes the feature matching algorithm based on SIFT.Among them,the k-d tree nearest neighbor algorithm is improved from the aspect of data structure,and index fields are added to make it have index information.When searching data,it can be more efficient and convenient,shorten thesearch time,and improve the search efficiency.At the same time,it avoids the matching of multiple feature points and the same feature point,reduces the search space and improves the accuracy.For RANSAC(Random Sample Consensus)algorithm,spatial constraints are added between matching feature points,and matching feature points are filtered.Because in the text image,the distance and angle between the correct matching feature points are approximately equal,in this paper,through the length and angle filtering,the wrong matching feature point pairs are removed,and the accuracy of the matching feature points is improved.