Motion-blurred SIFT Invariants Based On Sampling In Image Deformation Space And Optimization Searching

In the field of digital image processing, image matching is a very popular topic, and its algorithm efficiency has a direct impact on the accuracy of target recognition. SIFT(Scale-invariant feature transform) is a classical image matching operator, the operator has invariance of rotation, scale and illumination changes, and for affine transformation, changes of view also has certain stability, but SIFT does not adapt all types of deformation of images. Motion blur is a common type of image deformation, but result of SIFT matching is not satisfactory. Therefore, this paper presents a motion-blurred invariant SIFT algorithm that is based on sample matching in an image deformation reconstructed space. First, based on the mathematical model of motion blur deformation, we discretize the motion blurred parameter pair, and construct the space samples. Then, matching the images between the blurred image and the blurred samples. Finally, to find the optimal matching sample with maximal matching points. In order to improve the matching efficiency for the optimal sample, we used a univariate search technique combined with a variable step hill-climbing method. The motion blur deformation model has two parameters, so the reconstruction space is a two-dimensional space, univariate search technique is a good method to find the optimal solution for multidimensional space search. Combining the univariate search technique and the variable step hill-climbing method, greatly improve the efficiency of the algorithm.Due to the good ductility of SIFT, many scholars have made many improved algorithms to make SIFT have better matching performance. For example, ASIFT is a improved algorithm with better matching results. Compared with the SIFT algorithm, ASIFT has the completely affine invariance, but the matching effect for motion blurred deformation image is not better than that of affine deformation image, therefore the proposed method is used on ASIFT operator to improve matching ability of ASIFT under motion blur deformation.In this paper, a large number of image matching experiments are presented to verify that the proposed algorithm has a strong matching ability to the motion blurred image, the part of experiment completed the following contents: 1The comparison of motion-blurred image matching contrast between classical SIFT operator and proposed method. 2The comparison of motion-blurred image matching contrast between ASIFT operator and proposed method. 3The matching effect of life images. The experiment result show that the proposed method is superior on the matching number, matching effects, matching accuracy and practical application effect. It has reached the expected results in robustness and stability.