Study On Hidden Object Recognition Algorithm In Terahertz Human Body Image

The passive THz imaging system can detect metal and non-metallic materials by absorbing THz radiation form human body.However,passive THz imaging has a unique form of noise and is difficult to process with conventional methods.Therefore,it is important to study algorithms for identifying metallic and non-metallic objects hidden under clothes in THz human images.The main work is in following aspects:The original THz images grayscale range is between [0 3500],the grayscale mapping of the original data to [0 255],can simplify the data and increase the contrast of the gray scale of the detector.Because in the THz image,the contour of the human body is not obvious,and the foreground and background gray values are not much different.In the process of studying the algorithm,it is easily disturbed by the background information.For this situation,the thesis defines the customize template filtering,connected area method filtering,morphology opening and closing operation and the edge smoothing filtering processes the grayscale mapped THz image,extracts the human image and removes the disturbances such as background and other irrelevant information.The background elimination THz images' noise is complex and varied.The thesis removes the salt and pepper noise in the THz image and smooths other non-impulse noises by the adaptive median filter firstly.The detailed information in THz image is protected as much as possible,and then uses the theory of blind deconvolution to achieve the initial restoration of THz images.Considering that THz noise has some periodic characteristics,According to the characteristics of the THz image spectrogram to designing frequency domain filters.Compared with conventional low-pass filters,The notch band-stop filter designed in this thesis obtains a better denoising effect.Study the denoising effect of THz images by non-local mean method and 3D block matching method,improved Wavelet analysis and 3D block matching algorithm,experimental results show that the improved filtering algorithm has a certain effect.Combining subjective and objective denoising effect evaluation index,in order to design the optimal algorithm sequence of this thesis,the thesis uses the gray-scale mapping of original THz data firstly;after background processing;then adaptive median filtering;and then through the blind deconvolution operations on the degraded THz image of the initial restoration;afterwards in frequency by a trap band stop filtering;finally,uses the BM3 D denoise the image.The THz image processed by pseudo color is added to obtain a clear target THz image.The algorithm is tested by measured images to prove that it is effective.