Study On Local Invariant Features Based On SIFT

Local invariant descriptor technology is very popular in nowadays, which has been widely used in many fields, such as face detection, target recognition, character recognition, image mosaics and robot vision. In current image processing technologies, it is a hot research field. Among those local invariant feature algorithms, SIFT(Scaled invariant feature transform) is a milepost type algorithm because of its high stability and good discrimination. Although this technology has a history more than thirty years, local invariant feature algorithm still has more room for improvement. First of all, the study on point interference has not been completed. Secondly, the evaluation index of local invariant feature algorithm needs a more objective standard. Finally, the SIFT algorithm needs to improve its efficiency.First of all, this paper chooses the SIFT feature as an example and study on the anti-jamming properties of invariant feature points. This paper has found that the extraction positions fluctuate in a certain range. This paper analyses the fluctuation of feature points in noise conditions,fuzzy conditions,light transform conditions and all those disturbances which are the common interferences in actual operations to achieve the fluctuation range of different feature points.Secondly, to make improvements in the field of scale construction and descriptor generation of SIFT, this paper brings in the SURF(Speed-up Robust Feature) algorithm to detect features. The SURF algorithm uses a frame filter which is designed in different scales to keep the scale information to approximate the DoH(Determination of Hessian) operator. This greatly improves the efficiency of feature extraction. In the step of generating descriptors, this paper chooses the BRISK(Binary Robust Invariant Scalable descriptor Key points) descriptor to develop the spped of generating and matching.Again, this paper proposes an evaluation index based on the fluctuation characteristic of feature points which compares algorithms by the stable point ratio and fluctuation radius. This may provide a new evaluation idea. Based on the evaluation index and the traditional evaluation index, this paper compares different local invariant feature algorithms and proves that the BRISK-SURF algorithm has advantages in matching speed, stability and accuracy of extraction.Also, this paper applies the method of image registration. By using BRISK-SURF algorithm to descriptor generation and matching and eliminating error points, this paper realizes the precise image registration. Meanwhile, this paper studies the common methods of target tracking and proposes a target tracking algorithm which is simple and easy to realize. Those provides that the local invariant feature algorithm can be used in engineering application.Finally, the paper summarizes the main work and points out that the paper still has many improvement spaces. Firstly, the study of feature points’ fluctuation characteristic should be introduced into the rotating vector and more complex factors. Secondly, it is necessary to improve the performance of BRSIK-SURF algorithm. Lastly, it is necessary to improve the performance of target tracing.