Research On Image Registration And Fusion Algorithm Based On DM6467

In modern reconnaissance technology, extraction of the outstanding features of multi-source images becomes more and more important because of the fact that single-sensor imaging always has some defects. An ideal image with clearer objects in the dark and better details can be generated by fusing the registered multi-source images. This paper studies image registration and fusion algorithms based on TMS320DM6467hardware platform.Firstly, this paper introduces the overall hardware and software platform, the background and the significance of the project. And then a further introduction of the hardware platform is put forward. The function and performance of the central processing unit DM6467and the configuration of different work mode of the video decoder TVP5158are highlighted. Next, this paper studies various image registration algorithms and accomplishes interpolation method, corner-detection method, NCC method and RANSAC method using Matlab. After analyzing the advantages and disadvantages of each method, we choose manual registration method for the project; besides this paper specifies several fusion methods including weighted average method(WAM) and Laplace pyramid method and then simulates both methods with MATLAB. In order to form a better fusion image, we use different fusion rules for images on different pyramid level. Software realization is discussed in the last. We modify some related linux drivers and finish the program that satisfy all functions that the project requests. Finally we rewrite the image registration and fusion algorithms from MATLAB to C language and edit the configuration files of the software platform so as to port the algorithms to DM6467in the Codec Engine framework.