Research Of The High Dynamic Range Imaging Of Dynamic Scene Based On Double-Exposure Fusion

As the requirements for digital image shooting clarity and information are increasing,obtaining high dynamic range images have gradually become a research hotspot in the field of digital imaging.The high dynamic range images have been widely used in digital photography,remote sensing,security monitoring and plenty of other fields.Due to the sensitivity of the film's photosensitive elements,the noise of the shooting system and the capacity of the recording equipment,the digital photography systems can only record images with limited gray levels.This will cause the loss of the real scene information.In some serious circumstances,a large number of over-exposed and under-exposed areas will appear in the image,which will affect the visual perception effect of the human eye.Multi-exposure image fusion technology is a good solution for the above problems.The multi-exposure image fusion technology has become an important technology for obtaining high dynamic range images in recent years.Existing multi-exposure image fusion algorithms and systems would consume excessive time resolution and have a long calculation time.Meanwhile,the existing mature systems often suffer in the "ghost" phenomenon.Aiming at solving the above problems,this paper carries out the following work.Firstly,the existing multi-exposure fusion algorithm is reviewed.Based on the different types of imaging scenes,we classify the image fusion algorithms and ghost removal algorithms into static and dynamic scene algorithm.The existing mature systems are also reviewed.On the other hand,the multi-exposure fusion principle is discussed.Secondly,aiming at solving the problem that the brightness of the double-exposure image is too large,this thesis proposed an algorithm based on the Retinex theories to unify the brightness of the two frames.The algorithm averages the light layers and re-maps the reflection layer of the double-exposure image based on the principle of Gaussian function,which makes image pixels completely mapped.Considering that there may be both static and dynamic scenes in the one video sequence,an adaptive threshold selection algorithm is employed.To increase the universality of the algorithm,this thesis improves the algorithm for color images.Experimental results prove that the two-dimensional entropy,average gradient,the spatial frequency of the fusion image have reached 12.25,11.51 and 40.97,respectively.Compared with Mertens T's,Paul S's,Sen P's and Ma K's algorithms,these three parameters improve 3.81%,14.69%and 7.92%,respectively.Compared with the dynamic scene fusion algorithms proposed by Sen P et al.and Ma K et al.,the running time of the algorithm is increased by 58.5%on average.The proposed algorithm has lower operation time and higher advantages in terms of image information,sharpness and details.Thirdly,this thesis develops a high-dynamic-range video fusion camera based on a double-exposure model,which uses an FPGA to control the camera register and control the CCD chip to obtain a double exposure video image with continuous and alternating exposures.Proved by the double-exposure experiments on multiple scenes,the designed camera can capture the scene with a large brightness range and can reach half the frame rate of the original video sequence.