Date of Award
Doctor of Philosophy (PhD)
Vijayan K. Asari
Stephen A. Zahorian
Image enhancement techniques for visibility improvement of 8-bit color digital images based on spatial domain, wavelet transform domain, and multiple image fusion approaches are investigated in this dissertation research.
In the category of spatial domain approach, two enhancement algorithms are developed to deal with problems associated with images captured from scenes with high dynamic ranges. The first technique is based on an illuminance-reflectance (I-R) model of the scene irradiance. The dynamic range compression of the input image is achieved by a nonlinear transformation of the estimated illuminance based on a windowed inverse sigmoid transfer function. A single-scale neighborhood dependent contrast enhancement process is proposed to enhance the high frequency components of the illuminance, which compensates for the contrast degradation of the mid-tone frequency components caused by dynamic range compression. The intensity image obtained by integrating the enhanced illuminance and the extracted reflectance is then converted to a RGB color image through linear color restoration utilizing the color components of the original image. The second technique, named AINDANE, is a two step approach comprised of adaptive luminance enhancement and adaptive contrast enhancement. An image dependent nonlinear transfer function is designed for dynamic range compression and a multiscale image dependent neighborhood approach is developed for contrast enhancement. Real time processing of video streams is realized with the I-R model based technique due to its high speed processing capability while AINDANE produces higher quality enhanced images due to its multi-scale contrast enhancement property. Both the algorithms exhibit balanced luminance, contrast enhancement, higher robustness, and better color consistency when compared with conventional techniques.
In the transform domain approach, wavelet transform based image denoising and contrast enhancement algorithms are developed. The denoising is treated as a maximum a posteriori (MAP) estimator problem; a Bivariate probability density function model is introduced to explore the interlevel dependency among the wavelet coefficients. In addition, an approximate solution to the MAP estimation problem is proposed to avoid the use of complex iterative computations to find a numerical solution. This relatively low complexity image denoising algorithm implemented with dual-tree complex wavelet transform (DT-CWT) produces high quality denoised images.
"Multi-Modal Enhancement Techniques for Visibility Improvement of Digital Images"
(2005). Doctor of Philosophy (PhD), dissertation, Electrical/Computer Engineering, Old Dominion University, DOI: 10.25777/11b0-a981