Parametric blind deconvolution for passive millimeter wave images basing on image decomposition

Abstract

This paper deals with deconvolution problem for passive millimeter wave images with poor resolution and low SNR. A passive millimeter wave images super-resolution algorithm based on semi-blind deconvolution is put forward. The proposed method is based on two characteristic of imaging system. First, the PSF of imaging system is certain, can be modeled by parametric function. Second, the noise imposes different influence degree on the low frequency and high frequency parts of the pass-band of the image, the low frequency and high frequency part have high and low SNR, respectively.The image is decomposed using the bilateral filtering into low frequency base layer and high frequency detail layer. The base layer contains the large-scale structures and nearly frees with noise thus has the higher SNR, whereas the detail layer includes both small-scale details and noise and has the lower SNR. The base layer is restored by semi-blind deconvolution. The system PSF is modeled as a parametric Gaussian form. Edge structures information of the image is extracted basing on Mumford-Shah model and used to adjust the regularization term adaptively, and iterative method is used to estimate image and blurred kernel parameters. The detail layer is adaptively denoised by combining the joint bilateral filtering method and with edge preservation in the guidance of the base layer. Finally, the high resolution image is obtained by combining the base and detail layers. Comparative experimental results show that the proposed method can effectively suppress noise, reduce artifacts, and improve the spatial resolution.