Research Of Multi-Exposure Image Fusion Approaches Using Structural Patch And Pyramidal Decomposition

Due to the limited exposure range of the digital imaging device,it is difficult to get clear image for all objects in the same scene.Multi-exposure image fusion technology has been widely used in the field of computer vision.In general,multi-exposure image fusion may be from two aspects: pixel-level and patch-level.In this paper,two kinds of effective image fusion algorithms based on pixel level and patch-level are proposed.The main contents and innovations of this master's thesis can be summarized from the following aspects:1.This master's thesis briefly analyzes the background and current research status of this research field.The theoretical knowledge of multi-exposure image fusion is also introduced.2.Aiming at the limited of the digital imaging device,this master's thesis introduced a decomposition fusion algorithm based on image pyramid at pixel-level.Firstly,the algorithm blends multiple exposures,conducted by ordinary quality measures like saturation and contrast and well exposure.The resulting image is skipped the existing tone mapping physically operators,which avoids camera response curve calibration.Then the weight is selected by a simple power function that making the algorithm computationally efficient.Finally,multi-exposure images are fused into a high-quality result image with rich texture and color details.And enlarging the dynamic range of the scene for the final fusion of image performance.The algorithm produces a pleasing fusion image with a good color appearance and mainly texture details.3.In the patch-level,the proposed algorithm summarizes the experience of pixel level image fusion.Firstly,the multi-exposure images are decomposed into three different independent parts: contrast extraction,structure preservation and intensity adjustment.For structure preservation and intensity adjustment,which are very important,the master's thesis exploits three weight measurements,i.e.,local weight,global weight and saliency weight.From those weights,the final fusion image is guided not only by the exposure level of a single image but also the relative exposure level between different exposure images.After that,the three parts of the image are decomposed and weighted by the three channels of RGB,respectively.Finally,a desired patch is reconstructed and fed back to the fused image.Experimental results demonstrate that the proposed approach has better both in subjective and objective indexes.