A New Algorithm For Fast Image Stitching Based On BRISK

On the premise of guaranteeing the quality of images,we are able to obtain ultra-wide angle or 360 °panoramic images by image stitching.As a result,image stitching technology developed gradually.In recent years,image stitching is becoming one of the hot issues in computer vision.The technology is widely used in remote sensing images,medical images,virtual reality,video compression,video transmission,video retrieval,three-dimensional reconstruction,etc.This paper mainly studies the following three problems:(1)While extracting the features,the threshold of BRISK operator is constant.However,there exist some shortcoming in this process,such as the number of feature points is too large;some wrong feature points exist;etc.To solve these problems,this paper proposes a dynamic threshold to improve the operation method of BRISK operator.According to image inherent attributes,this new method adopts the between-cluster variance method to calculate the dynamic threshold.By experiments,we found that this method can extract more correct feature points of reasonable numbers with keeping the original speed.The operator can adapt to the image and the robustness is stronger than the original operator.(2)After the feature descriptor is generated,the initial matches between images will be accomplished.However,the initial matches have a large number of wrong matching pairs.If the purification process is carried out directly,the and the overall speed of stitching will be affected.To tackle these shortcomingscalculation cost is very large,this paper proposes to calculate a measure of similarity matching in the initial matching before removing the obvious mismatch according to the measure.This measure is calculated from the distance between the initial matching points.The experimental results show that the wrong matches can be removed on the premise of keeping the precision stable.Then,the calculation time of RANSAC algorithm to purify will be reduced.(3)The RANSAC algorithm chooses the matching point randomly to calculate the transformation matrix and iteratively obtains the best transformation matrix.But the cost of computing is expensive.If the matching point is not well chosen,it will be more difficult to find the best transformation matrix.In this paper,we propose a simplified RANSAC algorithm to find the best transformation matrix which is faster than the original algorithm.The main idea is as follows.Since the orignal RANSAC algonithm has the randomness of the selection of the matching point,we will sort from high to low based on the size of the matching metric.Then,we calculate the point according to the order.This method is able to avoid the first choice of false match points and find the best transformation matrix more quickly.The results show that a fast image stitching algorithm based on BRISK proposed in this paper satisfies the requirements of image stitching.