No abstract

No abstract

This paper investigates the use of neural networks for the direct estimation of image texture. Unlike previous approaches where networks are used to make decisions on feature vectors derived from traditional techniques, or where a network is trained to perform the function of a traditional technique, the proposed approach will use a network to directly model texture. The envisioned approaches to this method are described and the results of the preliminary l-dimensional tests are presented.

Remotely sensed bathymetry in the vicinity of Vieques Island and Ke( West, Florida is performed using Landsat Thematic Mapper imagery and a multiband model. Previous bathymetric algorithms using a single band or a ratio of bands assumed a constant bottom reflectance and thus required a bottom -type classification to isolate areas of uniform reflectance.The multiband model described in this paper does not require homogenous bottom -types and yields somewhat improved results over older methods.Depths to 16 m are measured with RMS residuals of less than 2 m. Results using other algorithms will be compared to the results from the multiband model. ABSTRACTRemotely sensed bathymetry in the vicinity of Vieques Island and Key West, Florida is performed using Landsat Thematic Mapper imagery and a multiband model. Previous bathymetric algorithms using a single band or a ratio of bands assumed a constant bottom reflectance and thus required a bottom-type classification to isolate areas of uniform reflectance. The multiband model described in this paper does not require homogenous bottom-types and yields somewhat improved results over older methods. Depths to 16 m are measured with RMS residuals of less than 2 m. Results using other algorithms will be compared to the results from the multiband model.

A technique for correcting for haze and sunglint in Landsat Thematic Mapper (TM) imagery in coastal regions has been developed and demonstrated using Gram-Schmidt orthogonalization of the band covariance matrix. This procedure is an adaptation of Wiener filtering and noise cancellation used in stochastic signal processing. Using a covariance matrix constructed from an over water portion of the image containing haze and sunglint pixels, a transfer function between infrared (IR) bands (e.g. TM 5) and visible bands (e.g. TM 2) is derived. This transfer function is then applied to the entire image and the visible band contribution predicted by the JR is subtracted from the measured visible signal, pixel by pixel. A comparison between images with and without haze of the same scene indicates that the procedure allows the observation of underwater features not previously visible.