Research On Methods Of Generating High Dynamic Range Image Based On Single Image

The dynamic range of the scene in the natures is very high,and the bright and dark regions often appear at the same time.However,the dynamic range perceived by human eyes exceeds by far that of the conventional sensor and the display devices.It leads that human can't obtain the real scene information due to the loss of detail and color distortion in the image.The high dynamic range(HDR)images not only maintain the HDR scenes,but also show that exquisite color information and clear brightness level,which are consistent with the Human Visual System(HVS).Thus,the HDR images acquisition based on multiple exposure images fusion is proposed in this paper.We present an exposure dependent curve to convert one LDR image to the pseudo-multiple-exposures.Then,the pseudo-exposures are fused into an HDR image by the proposed exposure fusion algorithm.The multi-exposure images acquisition includes the hardware-based and software-based methods.For the former,the image sequence is acquired by adjusting the aperture and shutter of the acquisition tool,then the final multi-exposure image sequence is registered.For the latter,the final multi-exposure image sequence is obtained by adjusting the brightness of the normal exposure.In this paper,some classical algorithms of these two methods are studied in details,and then the deficiencies are analyzed.An improved tone mapping algorithm based on Retinex is proposed to obtain the multiple exposure image sequences.Multi-exposure image sequence fusion algorithm includes block-based fusion and pyramid-based fusion algorithms.First,we introduce two block-based fusion algorithms,the fusion methods based on the principle of information entropy and of the local characteristics.Then,the fusion algorithms based on Laplacian pyramid and gradient pyramid are introduced,and the advantages and disadvantages of these several algorithms are analyzed as well.Finally,to restore the dynamic range of the real scene,an improved fusion algorithm based on gradient pyramid is proposed,and the Gaussian smoothing luminance is used as the local region quality measures.HDR image quality evaluation algorithm includes subjective evaluation and objective evaluation algorithms.The proposed algorithm is evaluated by the subjective and the objective quality evaluation.The former adopts the double excitation method,and the latter mainly uses the correlation entropy method,the dynamic range variable image quality evaluation and the HDR-VDP-2.The evaluation results show that the algorithm proposed in this paper has better performance and is more in line with human visual characteristics.