Design And Implementation Of Image Mosaic System Based On SIFT Algorithm

Image mosaic is an image processing technology which aims to stitch multipleimages that have partially overlaps of the same scenario into a large-size and seamlesspanoramic image. This technology mainly includes four aspects: image capturing, imagepreprocessing, image registration and image fusion. Among them image registration is thecore technique. The algorithm of feature point detecting and matching based on SIFTalgorithm is one of the most classic algorithms in the research field of image stitching. Butthere is a major problem in this algorithm that a mass of mismatching points which affectsthe result are still exist.According to the problem described above, a new algorithm aiming at purifying thematching points is proposed. This algorithm screened more accurate matching pointsamong the existing rough ones by threshold filtering methods. The main idea of the newalgorithm is introduced as follows. Firstly it finds the min value of similarit y distanceamong the rough matching points and obtains a threshold based on the value, and thenfilters out the matching points which have a larger similarity distance than the threshold.The rest of the matching points are the final points used for image registration. For thebrightness inconsistency problem in image stitching, a novel algorithm for adjustingbrightness of the original images is put forward. The algorithm can get a mean value bycalculating the difference of brightness between the two adjacent images which are usedfor image mosaic, and then the mean value can be used as compensation for adjusting thebrightness of those initial images. After the stitching process, the brightness of panoramicimage will be consistent on some.On the basis of such improved SIFT algorithm, an image mosaic system is designedand developed, which owns the functions such as importing source images, stitching anddisplaying. It’s proved that the number of mismatching points is effectively reduced byusing the threshold filtering method in matching feature points. Meanwhile by means ofbrightness adjustment before source images stitching, the flaws of visible seams andoverall brightness inconsistency have ironed out largely in the final panoramic image.