Research On Recognition Method Of Concealed Objects In Backscattering Body Image

Image recognition has been one of the most hot research fields in computergraphics. This paper mainly studies the recognition of backscattering body concealedobjects, includes: image preprocessing, feature extraction and the recognition ofconcealed objects.According to the characteristics of the human body image, firstly, we adopt acombination denoising method, which is composed of median filter and Gaussianfilter method. This method not only can remove the noise, but also can retain filteringimage edge information. Then, An multi-level segmentation method，which is basedon the global segmentation and the local segmentation, is proposed. The globalsegmentation separates the body and the concealed objects area from the CBS image,and the local segmentation segments the concealed objects from the body area.Experimental results show that the method can effectively segment the concealedobjects. After the analysis of the concealed objects characteristic, we design acombination feature extraction method, which is composed of the classical momentinvariants, geometric parameters and objects space characteristic value. We useprincipal component analysis method for multidimensional data dimension reductionprocess. Finally, The BP neural network recognition technology is adopted torecognize concealed metal objects of human body images and good experimentalresult is obtained.