Performance analysis of wavelet based restoration for passive millimeter-wave images

A poor inherent resolution capability of the passive millimeter-wave (PMMW) imaging becomes a problem in many applications. The need for efficient post-processing to achieve resolution improvement is being increasingly recognized. To obtain high- and super-resolution PMMW imaging, many restoration methods have been developed and evaluated. In this paper, two recent advanced wavelet based methods are discussed; Fourier-wavelet regularized deconvolution (ForWaRD) and multiscale entropy method. The ForWaRD is a linear deconvolution algorithm that performs noise regularization via scalar shrinkage in both the Fourier and wavelet domains. The ForWaRD has been reported to be efficient and applicable to all ill-conditioned deconvolution problems. The multiscale entropy method, which generalized the wavelet-regularized iterative methods, is advance of the maximum entropy method (MEM), which is more effective and leads to efficient restoration. These two methods have not been applied and analyzed in the PMMW images which were highly blurred and low signal to noise circumstance. We have studied the restoration performance of wavelet-based methods in the PMMW imaging comparing with particular reference to the Lorentzian method. The evaluation has been performed with actual radiometer imaging with the 94 GHz mechanically scanned radiometer as well as simulation. In the actual radiometer imaging, a simple blind restoration method was exploited with blur identification. To compare the restored image fidelity, objective and subjective criteria were used, and the super-resolution capability was also checked. Comparison of the linear and non-linear methods revealed the preferable bandwidth extension of the non-linear methods. In the non-linear methods, the multiscale entropy and Lorentzian, they showed their strength and weakness.