Panorama Image Mosaicing Research And Implementation

Panoramic image mosaic theory, techniques and methods is currently a hot topic inimage processing. It mainly solve the problem that due to the limited perspective of imagingequipment, the360-degree field of view image cannot be directly taken. This research canpromote the progress of land survey, medical image synthesis, panoramic images of militarycombat support systems, civil panoramic camera and other technologies. It has very greatpractical importance and significant engineering application perspective.In this paper, the study object is the image obtained by a circular motion of a singlecamera and multi-camera. The theory and methods of the panoramic image stitchingtechnology based on the characteristics is studied in-depth. For the current problem onpanoramic image stitching, a series of complete solutions are proposed. Image preprocessing,image registration, image location, bundle adjustment, photometric optimization, transitionalzone fusion and panoramic projection involved in image mosaic is analyzed in detail andspecific realization algorithm is given. Finally a panoramic image mosaic system is designedand implemented using C++language.For the radial distortion of the original image in the image pre-processing, parameters ofthe camera are obtained using Zhang zhengyou calibration method for imaging equipment;polynomial is used to approximate the radial distortion of imaging device with non-lineariterative estimation of polynomial coefficients to obtain the optimal distortion parameters;then the original image is corrected using the camera parameters and distortion parameters toobtain distortion-free images. The imaging device calibration and the correction of radialdistortion are effectively solved.In image registration, feature-based registration is researched deeply for matchingproblems of multiple images of the same scene obtained by different imaging devices, atdifferent times and at different angles. SIFT feature extraction methods are improved usingimage invariant moments and the characteristics of the image is obtained; then using Best BinFirst (BBF) search method, a collection of the early matches are obtained and an improvedRANSAC algorithm is given; finally adopting the projection transformation model of thewhole camera movement and the the matching set after purification, transformation matrix iscalculated and image registration is complete.For the cumulative error caused by image cascade transform and the problem ofdistortion and deformation involved in the projection (stitching) of all images on the panorama, bundle adjustment based on Levenberg-Marquardt (LM) algorithm effectivelyeliminates the accumulated error and deformation effect.For the problem that since images have different photometric parameters, the wholepanorama after stitching has uneven and unnatural brightness and color, a new photometricadjustment technique is proposed. This method optimally adjusts overall brightness and colordifferences of all the images and achieves a good image photometric adjustment effect.Image fusion is mainly used to solve the "seam-line" caused by local differences inbrightness in image stitching part and the "ghosting" caused by registration error inregistration and bundle adjustment. In this paper, the optimal stitch line is found using themethod based on distance transform. Then the wavelet multi-resolution analysis combinedwith cos (x) smooth function is adopted to achieve better stitching trace elimination effect.For the problem that panorama has single format and cannot meet the needs ofmulti-domain, six commonly used projection methods are discussed in this paper and givesthe output algorithm of panoramic image using these six common projection format.Finally, combined with the improved panoramic image stitching method, automaticpanoramic image mosaic system (automatic panoramic image mosaic system-APIMS) isresearched and realized programming in C++language. The system has the ability to stitchimages with overlapping regions from a small angle camera into high-quality360°panorama.The results of three group-experiment show that APIMS can achieve very versatile andhigh-quality panoramic image, which prove the accuracy and effectiveness of the proposedimproved panoramic image stitching method. A set of subjective and objective evaluation ofthe contrast experiments show that APIMS and the corresponding improved panoramic imagemosaic technique outperform than traditional stitching methods in time, effect and quality.Therefore, APIMS is suitable for not only mapping, medical imaging synthesis, virtual reality,computer vision and other civilian areas, but also military field of panoramic combat supportsystems such as tanks and submarines.