High Dynamic Range Image Construction and Noise Reduction Using Differently Exposed Images

Sammanfattning: This thesis discuss how to use mutual information in differently exposed images to improve noise reduction. It also investigates how one can create an high dynamic range (HDR) image from multiple differently exposed images taken with a handheld camera while simultaneously cope with the problems this approach introduces. The proposed method and workflow in this thesis is based on the non-local means algorithm that uses both image registration and a intensity transformation to takes advantage of the mutual information in the different images. It uses alpha expansion to select contiguous areas for the HDR construction and image blending to create seamless transitions between the images. The noise reduction algorithm shows better results with an intensity based noise level compared to a constant one. The methods used to take advantage of the mutual information are proven to be inadequate as the result for using noise reduction in a single image is shown to be just as good. The HDR image construction using both alpha expansion and image blending works well. For smaller movements the approach shows good result and even works as an anti-ghosting algorithm but for larger movements ghosting artifacts are introduced.