Compression and reconstruction of hadamard naturalness-preserving transform coded images
Rao Yarlagadda, C.R. Fore · 1999
Scope and method of study. This work investigates the use of the Hadamard Naturalness-Preserving Transform (HNPT) for image compression and reconstruction for use in situations where the communications channel results in severely degraded output images due to noise and/or jamming. An image coder is proposed using the HNPT along with a decoder-based restoration scheme based on convex projections techniques. The Kolmogorov-Smirnov test is used to determine goodness-of-fit between the histogram data and both the generalized Gaussian probability density function and a gamma density function. By using constraint sets based on the characteristics of the HNPT coefficients and on natural image characteristics, the reconstructed image is obtained by alternating projections onto the defined constraint sets. Findings and conclusions. Two sets of quantizers have been designed for use in encoding the HNPT coefficients. Both Lloyd-Max quantizers and uniform threshold quantizers are designed for the gamma distribution using shape parameters of either γ = 0.6 or γ = 0.5. These quantizers perform well in quantizing the high sequency coefficients. Using constraint sets based on the characteristics of the HNPT coefficients has proved successful in improving image quality even in example cases where thirty percent of the transform coefficients are lost. By enforcing consistency with known transform coefficients, corrupted coefficients were brought closer to an estimate of their original magnitude. An image restoration algorithm is developed which improved the quality of degraded HNPT coded images within 5 iterations.