Development of Multi-Exposure Image Fusion Technique

Abstract

Standard imaging devices are unable to encapsulate all the details of a natural scene as regular imaging sensors utilized in advanced cameras usually have exceptionally restricted catch go rather than the wide powerful scope of a characteristic scene. Subsequently, a solitary caught picture will in general lose a few subtleties in both over-exposed and underexposed sections. Although high dynamic range (HDR) gadgets can be used to alleviate this problem, but these are very expensive. Hence, in order to encapsulate subtleties of a complete scene, wide number of multi-exposure fusion (MEF) techniques have been put forward, yet there is a need for efficient, accurate and robust MEF technique. In this thesis, an effective MEF scheme is proposed, which takes a combination of both pixelwise and patch-wise methods to extract weight maps utilizing dense scale invariant feature transform (DSIFT) and three conceptually autonomous components; signal structure, loca mean intensity and signal strength respectively. Moreover, the fusion of all the extracted weight maps yield an output image which is of enhanced quality, does not require empirical settings of parameters and anticipated to be more enlightening and outwardly engaging than any of the source pictures. Proposed MEF can be used in security and surveillance purposes as well as in biometric recognition. Moreover, maintains colour balance and natural smoothness of a scene, noise and halo artifacts are also avoided in proposed scheme. Visual and objective analysis with best in class existing technique has been performed to verify the significance of proposed technique using ten sets of multi-exposure images and two fused image quality models respectively.