| The passive millimeter wave(PMMW) imaging technology is one of the effective means to detect dangerous objects concealed under clothing. Due to the limitations of the available hardware and the inherent physical properties of PMMW imaging systems, the acquired images often suffer from serious noise pollution and some distortion. By analyzing the main influence factors of the PMMW imaging system as well as the inherent characteristics of PMMW images, this dissertation develops an improved denoising algorithm, a fusion algorithm in wavelet domain, and an improved nonlocal means denoising algorithm. In addition, a novel image segmentation algorithm is proposed and applied to PMMW imaging system. The main contributions in this dissertation are summarized as follows:(1) Through the establishment of radiation temperature model, we analyze the relationships between the radiation temperature difference of body and concealed objects and the transmission of clothing, and between the emissivity and the transmission of the concealed object. For some items that cannot be detected by PMMW imaging detection system, the feasibility of increasing their absorption performance by adding very small amounts of microwave absorbing materials into such items is researched. Then the influence of the modified absorbing material on the microwave absorbing properties of samples is analyzed.(2) The wavelet denoising and fusion algorithms are applied to PMMW image information processing. To address the problem that wavelet denoising is vulnerable to interference of the discontinuous points, an improved translation invariant Bayesian decision wavelet threshold denoising algorithm is proposed. The operation of symmetric periodic extension eliminates the negative influence of discontinuous points on PMMW image boundary to a certain extent. In terms of wavelet fusion, a fusion strategy combining structural similarity(SSIM) and image edge energy is proposed, which are capable of inhibiting the ringing effect caused by the single use of SSIM fusion criterion and thus improving the quality of the fusion image.(3) PMMW images are composed of a small amount of piecewise smooth image blocks, which contain lots of redundant information. According to such characteristic, a modified pre-selection based nonlocal means(MPSNL) denoising algorithm is developed. In the MPSNL method, PMMW image is first divided into three kinds of regions by using the singular value decomposition: smooth region, transition region and edge region. Then three kinds of nonlocal denoising rules suitable for PMMW images are proposed. The MPSNL algorithm overcomes some disadvantages of traditional nonlocal means(i.e. NL-Means) algorithm such as heavy computational load and low utilization rate of the image feature.(4) Image segmentation plays an important role in identification of concealed objects based on PMMW technique. Therefore, a precise and fast segmentation algorithm suitable for PMMW images is necessary. In our work, an adaptive parameter initialization Gaussian mixture model algorithm(APIGMM) is proposed to extract histogram feature. First, the gray level distribution range of edge, called confidence interval, is obtained from the global perspective. Then the Canny operator is employed to extract the image edge from the regional perspective. Finally, only the edge in the confidence interval is preserved. The experimental results demonstrate that the proposed algorithm can achieve fast and accurate segmentation of concealed objects.Based on the analysis of the characteristics of PMMW images, a series of improvements are made to different types of image processing algorithms. In particular, the image denoising and fusion algorithms in wavelet domain exhibit certain universality. Apart from the PMMW imaging, they also can be applied to other types of image information processing systems. The MPSNL denoising algorithm and improved image segmentation algorithm are applicable to PMMW image information processing.
|
PDF Full Text Request