Multi-exposure Image Fusion Based On Information Theoretical Measure

With the continuous improvement of the camera functions of mobile devices such as mobile phones,photography is becoming more and more convenient and fast,and with the rapid development of the Internet,digital images are almost everywhere,and the content and performance of images are constantly enriched,making people constantly pursue technological perfection.However,the range of colors that digital images can represent is limited.For a normal digital camera,details in brighter and darker regions can not be fully captured in one exposure,and rich details can't be displayed completely.Therefore,people hope to get high dynamic range(HDR)images that approximate real scenes.Currently,image fusion is a very popular algorithm for acquiring HDR images instead of high-performance devices.In this paper,we proposed a multi-exposure image fusion method based on information channel.For two different exposed images,information channel from one image to the other is constructed,based on which we have proposed an information measure to estimate the pixel-wise information,then the weight maps of the two images can be generalized by normalizing the information.For the fusion where the input images are more than two,we build the information channels between each pair of images.Considering other images have different influence on the information of one image,so that we compute gaussian function of luminance difference to get the relevance of every two images.Then we can get weight maps of each image by wighting average the relevant information from other images.In the other hand,to avoid the noises of direct fusion,instead of blending images by their weight maps directly,we conduct Laplace pyramid decomposition for each source image,and fusion by Gaussian-filtered weight maps is carried out for each layer.In the end,the final image is reversely reconstucted by fused images at each scale.In this paper,we employed information theory to measure image information,and applied it into multi-exposure image fusion.The proposed multi-exposre fusion algorithm is easy to implement,and delivers results which are competitive with state-of-art methods.