Spatial Enhancement of Tir Aster Data via Vnir Images and Generalized Laplacian Decomposition

Image fusion aims at the exploitation of the information conveyed by data acquired by different imaging sensors. A notable application is merging images acquired from space by panchromatic and multior hyper-spectral sensors that exhibit complementary spatial and spectral resolution. Multiresolution analysis has been recognized efficient for image fusion. The Generalized Laplacian Pyramid (GLP), in particular, has been proven as the most efficient scheme due to its capability of managing images whose scale ratios are fractional numbers (non-dyadic data) and to its simple and easy implementation. Data merge based on multiresolution analysis, however, requires the definition of a model establishing how the missing spatial details to be injected into the multispectral bands are extracted from the panchromatic image. The model can be global over the whole image or depend on the local space-spectral context. This paper reports results on the fusion of Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data. Each of the five thermal infrared (TIR) images (90m) is merged with the most correlated visible-near infrared (VNIR) image (15m). Due to the 6:1 scale ratio, the GLP has been utilized. The injection of spatial details has been ruled by means of the Spectral Distortion Minimizing (SDM) model that minimises the spectral distortion between the resampled and fused images. Notwithstanding the lack of a spectral overlap between the VNIR and the TIR bands, experimental results show that the fused images keep their spectral characteristics while the spatial resolution is enhanced. INTRODUCTION Space-borne imaging sensors allow data acquisition of the Earth surface on a routine basis. Multispectral (MS) observations, however, exhibit ground resolutions that may be inadequate to specific identification tasks. As in the case of satellite imagers like Ikonos, QuickBird and SPOT 5, which make available very high resolution MS and panchromatic (P) data, image fusion can be conceived also for ASTER TIR data, whose resolution is not sufficient for many application tasks: the VNIR ASTER bands could be the source of the enhancing spatial details. Data fusion techniques take advantage of the complementary spatial/spectral resolution characteristics of imaging sensors to spatially enhance the acquired images. This specific aspect of data fusion is often referred to as data merge (i) or band sharpening (ii). In fact, in the case of P and MS data, the P band is acquired with the maximum resolution allowed by the sensor, while the MS bands are usually acquired at a coarser resolution, typically, two or four times lower. Once received, the P image may be merged with the MS images to enhance their spatial resolution. Since the pioneering high-pass filtering (HPF) technique (iii), fusion methods based on injecting high-frequency components into resampled versions of the MS data have demonstrated a superior performance (iv). HPF basically consists of an addition of spatial details, into a bicubicallyresampled version of the low-resolution MS image. Such details are obtained as the difference between the P image and its low-pass version achieved through a simple local pixel averaging. Later efforts take advantage from an underlying multiresolution analysis, by employing the discrete wavelet transform (DWT) (v), rational filter banks (vi), Laplacian pyramids (LP) (vii) and morphological pyramids (viii). Although seldom explicitly addressed by most of the literature, the rationale of high-pass detail injection as a spatial frequency spectrum substitution was formally developed in a multiresolution framework as an outcome of filter-banks theory (ix). © EARSeL and Warsaw University, Warsaw 2005. Proceedings of 4th EARSeL Workshop on Imaging Spectroscopy. New quality in environmental studies. Zagajewski B., Sobczak M., Wrzesień M., (eds) The DWT has been widely employed for remote sensing data fusion (x). According to the basic DWT merging scheme (xi), couples of sub-bands of corresponding frequency content are combined together. The merged image is synthesised by taking the inverse transform. Schemes based on the “à trous” wavelet algorithm were also recently proposed (xii,xiii). Unlike the DWT, which is critically sub-sampled, the “à trous” wavelet and the LP are over-sampled. The miss of the decimation step allows an image to be decomposed into nearly disjointed band-pass channels in the spatial frequency domain, without losing the spatial connectivity of its high-pass details, i.e. edges and textures. Starting from a synthetic yet comprehensive review of wavelet analysis, advantages for image merging of redundant multiresolution decompositions have been recently demonstrated (xiv). Advantages are found also for already established data fusion approaches once formulated in an undecimated multiresolution framework (xv). Furthermore, the LP can be easily generalised (GLP) to deal with scales whose ratios are whatsoever integer or even fractional numbers (vi). Data fusion (merge) based on multiresolution analysis, however, requires the definition of a model describing how the missing high-pass information to be injected into the resampled MS bands is extracted from the P band (iv). The model can be global over the whole image or depend on local context, either spectral (xvi,xvii), or jointly spectral and spatial (xiv,xviii). The goal is to obtain fused bands as similar as possible to what the MS sensor would image at the resolution of the P band. Preservation of spectral information, regarded as changes across spectral bands, or equivalently as colour hues in the composite representation of three bands at time, must be guaranteed after spatial enhancement. Hence, methods were developed based on the following steps: (a) transformation of the spectral bands, resampled at the scale of the P image, into Intensity-Hue-Saturation (IHS) coordinates, (b) replacement of the smooth I component with the sharp P image, (c) inverse transformation to the original spectral domain. IHS fusion methods (xix), however, may introduce severe radiometric distortions (e.g. bias in local mean) in the sharpened MS bands, due to the lowpass component of the P image that affects the fused product. To overcome such inconveniences, IHS fusion was either extended into a multiresolution framework (only details of the I component are replaced with those of P) (xii,xiii), or analogously the smooth I component was sharpened by modulating each pixel by the ratio of P to its low-pass version (xx). The latter procedure may introduce radiometric inaccuracies in textured areas, due to statistical instabilities and the bias of the rational term. All the above methods are used in the case of exactly three spectral bands. When the number of components is larger, IHS methods are applied to three bands at a time. Although IHS-based methods were specifically developed to preserve spectral information, the problem of spectral distortion was never explicitly considered, when the injection model was setup. Spectral distortion can be measured, regardless of the number of spectral components, as the absolute angle between a pixel vector in the true and in the fused MS data. Such an angle should be lower than or equal to that measured between the expanded original low-resolution MS data and the true high-resolution MS data, when available (e.g. in simulation carried out on spatially degraded images). The case of equality means that the fusion algorithm has preserved the available spectral information (SDM model) with the benefit of a spatial enhancement. In this work, the GLP fusion scheme (vii) is utilized for the fusion of TIR ASTER images by injecting spatial details extracted from VNIR bands, the injection being ruled by the SDM model. After reporting the definition of GLP, several fusion models embedded in the GLP scheme are reported. ASTER data characteristics are then examined in order to adapt the fusion algorithm to such data. Results are then reported and discussed. Eventually, conclusions are drawn.