Study On CCD Image Mosaic Techniques Based On Wavelet Transform

Image stitching algorithms get more and more attention in image processing. Practicability of the algorithms is very good.They are widely used in the fields,such as the synthesis of panoramic images,the processing of the earth satellite photos and medical images,and so on. Yet this paper mainly researches the CCD image stitching.Firstly,this paper generalizes and summarizes the image stitching techniques.And introduces some research results of the current image stitching technology.And summarizes the general rules of image stitching technology.Then wavelet transform theory is given a systematic introduction which focuses on the two-dimensional wavelet transform.In general,the image mosaic algorithms can be divided into two categories:one is based on the regions, and the other is based on the characteristics. The advantage of region-based algorithms is more intuitive.But the one drawback is the large amount of computation and is only suitable for the simple images.And the other one is that the noise of the images has apparent influence on the effects of mosaic.Because region-based matching algorithms directly make use of the correlation between pixels of overlapping region of the reference image and the sensed image.The ratio algorithm of matching is used in chapter4,image mosaics based on region.In the overlapping region of every image,the template is the ratio of two colums pixels which have certain distance.Then the ratio algorithm is emulated.By comparing the results of the simulation,we can conclude:ratio algorithm is not suitable for the approximative images,it will cause stiching errors;but it is suitable for closer and not close images.Wavelet transform is used in the algorithm,and the emulation results are given.The next chapter finishes feature matching algorithm based on the wavelet transform modulus maxima. Firstly it gives feature matching process. Since the maximum of the wavelet transform modulus has the information of image singularity points, they can be feature points.One modulus maxima point is not less than the neighborhood points and rigorously greater than at least one of the points above.It is correspondingly on the edge of the image.Before calculating the wavelet transform modulus maxima,the wavelet transform modulus must be got.This paper gives the emulation results of wavelet transform modulus of the sensed image and reference image.And gives the emulation results of wavelet transform modulus maxima when thesholds are diffenrent.Then gives the relation between mosaic speed and the number of characteristic points.Correlation algorithms are used in feature point matching.This paper gives the number of matching feature point pairs when correlation coefficient threshhold is 0.9.The global transformation model is used for forward map(mosaic) from reference image to sensed image.The result of map is given,and it has holes.The same model is used for backward map(mosaic) from sensed image to reference image.The result has not holes.Chapter6 of this paper is the conclusion and outlook.It elaborates the innovations and inadequacies of this paper,and looks into the future of the image mosaic techniques.