Focal Stack Image Fusion Using Image Volume Reconstruction

Abstract

Traditional camera can’t take photos with large depth of field (DOF) because of the lens limits. The region out of DOF often gets blurred due to narrow width of depth. Focal stack photography take a set of (three and above) pictures viewing the same scene with different focus point and fusion the pictures as one using computational algorithms in order to have all-in-focus photos. Some state-of-art focal stack algorithms seek the best-in-focus point among these stack photos and collect them into a new all-in-focus image. This method is derived from the assumption that at least one photo in the stack has the clear vision of the point of the scene. But what if none of these points are in focus because unfortunately focus point falls out of the object of the scene? Focal stack images not only provide us several pictures independently but also partial information about how the object is imaging at a certain depth. With this information, we seek a space called ‘image volume’ to search the best in-focus depth beyond the two dimensional images to avoid the failure mentioned above. We work through the whole scene space and find the accurate depth of every point in the scene. We experiment some pictures with focal stack example and show better results than traditional methods.