Design And Implementation Of Image Registration System Based On SIFT

Image registration technique is an important research direction in the field of computer vision.With the continuous expansion of the application field and the improvement of registration accuracy,the image registration technology developed on the PC has been unable to meet the current demand.In view of this,makes some adjustments to the SIFT algorithm,and proposes a design scheme of image registration system based on FPGA and DSP in this paper,then make the hardware platform and software development of the system.This paper mainly focuses on the basic theory and system implementation of image registration.According to the invariance of SIFT(Scale Invariant Feature Transform)features and good parallelism,make full use of advantages of Canny algorithm to eliminate the residual of edge points after SIFT,and combine with the RANSAC algorithm to correct the mismatching points in the initial match.And using OpenCV simulation analysis on the PC,the simulation results show that the algorithm has good effect with image registration.For image which the size of 640*480 and has complex edges,The algorithm can reduce the feature points about fifteen percent and can effectively improve the post-processing speed and noise immunity.Secondly,using FPGA and DSP as the core to build the image registration system hardware platform,do a simple introduction for the CMOS sensor,SDRAM memory,DAC video and other devices to;The FPGA module logic control and the application program of DSP image registration algorithm in the system are programmed by the modularization idea,through the EMIFA interface to read and write the external FIFO,which develop nuclear data communication interface unit and realize collaborative processing between FPGA and DSP.This paper describes the design process of the system from the aspects of the wholesystem,hardware platform,logic control and algorithm realization.Using ModelSim,ChipScope,CCS and other tools to modify and debug each module after the completion of hardware platform and logical algorithm development.Finally,the experimental results verify that the system designed in this paper can complete the image registration process.And the system has great application value in medical instruments,image mosaicking and other fields.