Dynamic Range Extension Based On Multiple Exposure Fusion And Ghost Removal

Many real-world scenes contain a wealth of contrast, brightness and color information, the dynamic range is far beyond the capability of conventional digital cameras. When taking photos of high dynamic range scenes, one can select the brightness range of interest by adjusting exposure. However, the single image captured by digital cameras will always contain dark or saturated regions, with many details lost. As the development of computer science and digital image processing technology, the digital image is applied to more and more areas. This also put forward higher requirements on the dynamic range of digital imaging system. Using a series of low dynamic range images with different exposures to generate a high dynamic range image and enhance the detail of image is an effective means to extend the dynamic range of imaging system. It is of great research value and scientific significance.In order to extend imaging dynamic range, we introduce the imaging process of common digital cameras, analyze the theory of dynamic range extension using multiple exposure images. A typical camera response function solving and high dynamic range image synthesis method is analyzed in detail, and methods about multiple exposure images matching and tone mapping are discussed and validated by experiments.For static scenes, the author proposes an efficient multi-exposure image fusion method. Firstly, we propose an evaluation method of well-exposure combining the scene brightness information, using the Gaussian mathematical model to calculate the evaluation value. The evaluation method is then used in image block area to calculate the weight value of each pixel. The multiple exposure image sequence is finally fused to a local adaptive exposure high dynamic range image, to achieve dynamic range expansion. Experimental and comparative results show that this method is efficient, the generated image has a greater ability to perform the texture detail and color information, and has a higher entropy.For dynamic scenes, a novel and effective ghosting artifact detection and removal methods is proposed. We use the super-pixel segmentation method to extract contours of the scene. Every input image is divided into image blocks comprising the local properties of the scene. Ghost detection is performed in each sub-block with the reference image, via zero mean normalized cross-correlation. Finally, the ghost super-pixels are excluded by adjusting the weight map for fusion.