Research on SIFT Algorithm Base on Complex Environment
Xiangying Dang,Dan Li
DOI: https://doi.org/10.2991/ISRME-15.2015.138
2015-04-01
Abstract:A new algorithm of feature matching was presented by this paper. In the new algorithm, Curvelet first used to denoise, then RANSAC optimized SIFT algorithm used to stitch images. Experiments show that the new algorithm is better to remove error matching points. It improved the accuracy of local invariant feature matching. New algorithm has a good anti-noise and lighting capabilities. It has strong adaptability to scale changes, the algorithm is robust, and has better effect on matching and stitching. Introduction Firstly, the paper used Curvelet algorithm to remove image noise, then used SIFT scale invariant feature vectors to match features. The improved Ransac method was used to remove the mismatch. The new algorithm is adapted to the complex environment with noise, light changes, blur and rotation problems.It is robust and accuracy while ensuring the matching efficiency. Image Preprocessing Due to the problems of complex environmental monitoring video has noise, low resolution. In order to accurately extract the feature points, need image preprocessing to remove noise. Wavelet has good time-domain and frequency-domain analysis capabilities, many scholars have proposed a method based on wavelet denoising, such as nonlinear thresholding wavelet transform, translation invariant denoising, etc. However, the two-dimensional separable wavelet has only a limited level, and diagonal direction, because of lack of multi-direction make it can not represent a line of exotic, bizarre face high dimensional function. Based on wavelet theory, E.J.Candes and D.L.Donoho established a ridge wave transformation from 1998 to 1999 [1]. Ridge wave uses a wavelet function by adding characterized direction parameters. It has a strong direction selection and the ability to identify and can effectively represent the direction singular characteristic in signal. To further indicate the multidimensional signal general curved singularities, and the Curvelet methods [2] can use the multi-scale local linear approximation of the curve. Noise and edge information can be well separated, while keeping an edge, the noise suppression achieved good results. This paper will apply the second generation Curvelct transform in image denoising, it will have higher image SNR and the image will display more clearly. A. Discrete Curvelet 1 2 [ , ] f t t , 1 0 t ≤ , 2 t n ≤ is the input function,Curvelet transformation [3] can be expressed as: 1 2 1 2 , , 1 2 0 , c ( , , ) : [ , ] [ , ] D j l k t t n j l k f t t t t j ≤ 0, it has
Computer Science