Research On Automatic Registration And Splicing Algorithm Of Leather Cross-section Image

As a significant subject of Computer Science,Image Processing Technology(IPT)has been widely and successfully used in various application fields,such as remote sensing technology,medical science,3D image reconstruction and so on.The kernel theory of IPT is to convert images into digital matrices,and utilize relevant algorithms to study those matrices subsequently.This paper achieved feature-based automatic image matching by using related matching algorithms,and further gained the better robustness by comparisons and analysis of above algorithms in two dimensions: the horizontal stitching of leather fiber images(panoramic image registration and mosaic technology)and the longitudinal registration of images(sequence image segmentation).There are two approaches in this paper to actualize the image registration.The first one is based on 2D wavelet transform and SIFT algorithm,the key steps are as follows.To eliminate the edge of the image,we use wavelet transform to preprocess the image;and then the SIFT algorithm is used to match the image.The second approach jointed genetic algorithm(GA)and SURF algorithm to optimize image processing.The advantage of this strategy is that the genetic algorithm could determine the threshold automatically when SURF finding the match threshold.Accordingly,the optimum solution of bound space is easier to be selected.Experimental results showed that the traditional SIFT algorithm has obvious advantages for the same layer image matching of leather fiber,but weak in the sequential image matching and the correct rates.After wavelet transform image processing,SIFT algorithm not only maintains the stability of itself,but also make the wavelet denoising to maximum advantage during matching process.As a result,the matching speed and correct rate are obviously improved.Meanwhile,our research reveals that the SURF algorithm performs better in high brightness.We proposed a GA-based threshold selection method to optimize SURF algorithm,resulting in huge count increase of matching points and 99% correct rate or above.In sum,this paper has great significance for the three-dimensional reconstruction of leather and the research on cross section of leather fiber..