Research On Feature Points Detection And Matching Algorithm For Tree Images

The feature point extraction and matching of the tree images can provide the most direct data base for the subsequent construction and visualization of the three-dimensional tree.In this paper,analyze the traditional Harris algorithm,SUSAN algorithm and SIFT algorithm of image feature point extraction technology,at the same time,the normalized cross correlation algorithm,sequential similarity detection algorithm and mutual information algorithm of image matching algorithms are carried on the detailed elaboration,select the appropriate method to deal with the tree images,and in view of the shortcomings of the algorithm,the main work done in the following two aspects:At the aspect of feature detection,feature points that are detected by using the traditional Harris algorithm are throughout the whole image and the algorithm is sensitive to scale change,this paper puts forward the multi-scale Harris feature detection algorithm based on saliency region detection to solve this problem.First of all,the saliency region detection algorithm is used to construct image significant figure,segment the resulting image by threshold to obtain a binary map,and take operation of erosion and dilation to optimize,and the final target segmentation region as a candidate region;Secondly,introduce integral factor and differential factor to achieve Harris algorithm using multi-scale calculation,at the same time,the circular template is used to suppress the non-maximal value of the corner response function,to detect the feature points of the image and mark it to realize the precise positioning of the feature points.To deal with the tree images,compared with the traditional Harris algorithm,this algorithm makes less number of feature points,at the same time without changing any parameters,for image rotation,can reduce the difference of feature point extraction,to lay a good foundation for the subsequent image matching and reconstruction.At the aspect of image matching,this paper proposes a fast NCC image matching algorithm based on wavelet pyramid search strategy for the problems that the traditional normalized cross correlation algorithm(NCC)is computationally intensive,the operation speed is slow and the correct rate is not very high.Firstly,wavelet pyramid is applied as the matching strategy,and using wavelet transform constructs an image pyramid structure with low to high resolution,the hierarchical matching can be achieved feature location from coarse to fine;at the same time,based on the traditional normalized cross correlation algorithm,the algorithm in this paper references three sum tables to calculate the mean of the image,the variance of the image and the cross correlation between the images,using a subtraction method instead of multiplication to reduce computational complexity and the amount of calculation in the matching process.To deal with the tree images,compared with the traditional NCC algorithm,this algorithm can get less error matching,matching points connection is better,and the time is less,which provides a stable data base for reconstructing the three-dimensional model of the tree images.