Pixel - Level Fusion of Image Sequences
using Wavelet Frames1
Oliver Rockinger
Daimler Benz AG, Systems Technology Research
Intelligent Systems Group
Alt Moabit 96 A
10559 Berlin
Germany
rockinger@DBresearch-berlin.de
Abstract
In this paper we propose a novel approach to the pixel level fusion of spatially registered image sequences. This fusion method incorporates a shift invariant
extension of the discrete Wavelet Transform, based on the concept of Wavelet
Frames which yields an overcomplete signal representation. The advantage of the
proposed fusion method is the improved temporal stability and consistency of the
fused image sequence. We show examples of the application of the Wavelet
Frame fusion scheme on both real world images and image sequences.
I Introduction
By the development of new imaging sensors arises the need of a meaningful combination of all employed imaging sensors. This problem is addressed by image fusion. The actual fusion process can take place at different levels of information representation, a generic categorization is to consider the different levels as signal, pixel, feature and symbolic level [3].
In the following, we discuss the pixel level fusion process. To perform pixel level fusion successfully, all input images must be exactly spatially registered, i.e. the pixel positions of all input frames must correspond to the same location in real world. To date, the result of pixel level image fusion is considered primarily to be presented to the human observer, especially in image sequence fusion. A possible application is the fusion of infrared and visible images obtained by an airborne sensor platform to aid a pilot navigate in poor weather conditions or darkness.
In case of pixel level fusion, some generic requirements can be imposed on the fusion result: The fusion process should preserve all relevant information of the input imagery in the composite image, while suppressing irrelevant image parts and noise in the fusion result. The fusion scheme should not introduce any artifacts or inconsistencies which would distract the human observer or following processing stages. In image sequence fusion arise the additional problems of temporal stability and consistency of the fused image sequence.
The paper is organized as follows: In section II we briefly summarize the Discrete Wavelet Transform and the extended concept of Discrete Wavelet Frames. In Section III the Wavelet Frame 1 In: Proceedings of the 16th Leeds Applied Shape Research workshop, Leeds University Press, 1996
