| Image fusion is widely used in various fields as an important branch of multi-sensordata fusion. Image fusion is a progress which is processing image data on the same targetcollected from multi-channel source, extracting enabling information from each channel,finally generating high-quality images. pixel-level image fusion is the basis of all levels, itretains much of the original information of the scene, Provides richer, more accurate andmore reliable information which other integration levels can not provide.This paper has studied the pixel-level image fusion algorithm, and discussing severalaspects of the algorithm, such as the characteristics, complexity, integration indicators. Ithas wrote code to simulating both types of algorithm in the MATLAB platform, Imagefusion effects were evaluated based on the subjective and objective indicators of fusionevaluation on the basis of the simulation algorithm. The visible and medium-wave infraredremote sensing images were laplacian pyramid, contrast pyramid, gradient pyramidtransformed respectively, fusing the two images using coefficient fusion rules whichselected the larger pixel values, larger contrast values, coefficient fusion rules respectively,Finally, the resulting images obtained by the three fusion methods were compared byobjective evaluation such as entropy, correlation coefficient and spatial frequency, theresults show that the laplacian pyramid image fusion method is the most simple, and it hasgot the best results.Because GPU can process large amounts of data concurrently, and multi-scale imagefusion algorithm has higher parallelism, this paper uses GPU to accelerate laplacianpyramid image fusion algorithm. In Microsoft visual stidio2010, CUDA5.5environment, ithas realized the integration of different image sizes, fusion results achieved the expectedresults, writing the time testing program, the core algorithm processing time has got a goodspeedup compared with the CPU, and it has achieved real-time image processing. |
PDF Full Text Request