ANALYSIS OF IMAGE ENHANCEMENT ALGORITHMS FOR HYPERSPECTRAL IMAGES

Abstract

This thesis presents an application of image enhancement techniques for color and panchromatic imagery to hyperspectral imagery. In this thesis, a combination of previously used algorithms for multi-channel images are used in a novel way to incorporate multiple bands within a single hyperspectral image. The steps of the image enhancement include image degradation, image correlation grouping, low-resolution image fusion, and fused image interpolation. Image degradation is accomplished through a Gaussian noise addition in each band along with image down-sampling. Image grouping is done through the use of two-dimensional correlation coefficients to match bands within the hyperspectral image. For image fusion, a discrete wavelet frame transform (DWFT) is used. For the interpolation, three methods are used to increase the resolution of the image: linear minimum mean squared error (LMMSE), a maximum entropy algorithm, and a regularized algorithm. These algorithms are then used in combination with a principal component analysis (PCA). The use of PCA is used for data compression. This saves time at the expense of increasing the error between the true image and the estimated hyperspectral image after PCA. Finally, a cost function is used to find the optimal level of compression to minimize the error while also decreasing computational time.