Research On Key Technologies Of Contourlet Based Panoramic Image

Contourlet transform is a new multi-scale, multi-resolution analysis tool. This paperstudied on the theory of Contourlet transform.According to the practical applicationrequirements of characteristics of data, much details in complex images in environmentalmonitoring and tracking of moving targets,In order to fully protect the feature information,saving the hardware resource and improve the real time of the system, The specific techniquessuch as Image coding,encoding of image, edge detection, segmentation and compressedsensing reconstruction are researched in depth. The paper presents several key technologiesexisting in the transform processing of panoramic images by Contourlet, and achieved goodresults.Firstly, this paper improves the SPIHT coding method that is wavelet based Contourlettransform. Though Contourlet transform can meet the direction of the problem, the directionalwavelet transform not enough to effectively represent the image texture and contour, thetransform redundant and the low efficiency of the direct coding still exists. The wavelet basedContourlet transform is non-redundant transform, and the sub-band decomposition direction isflexibility, but the traditional wavelet-based SPIHT algorithm by Contourlet Transform doesnot consider the relationship between the low-frequency sub-band and high frequencysub-band, which only looks for the relationship among the high-frequency sub-band. As itneeds to replace the transformed coefficients, all the process will certainly affect the qualityand efficiency of the coding. To solve the above issues, this paper proposes the idea ofconstructing a virtual low-frequency, which means by constructing a relationship between thelow-frequency sub-band and high frequency sub-band with the virtual low-frequency willmake a higher structure of the direction of the tree and better compression. It provides aneffective method to solve the problem of large panoramic image information, highredundancy, the unfavorable storage, processing and transmission.Secondly, this paper proposes dual-threshold edge detection algorithm based onnon-sampling Contourlet Transform. Because of the translational invariance makes thedetected edge location accuracy, the traditional edge detection results which based onnon-Contourlet Transform, exist pseudo-edge. This is due to the improper selection of thefixed threshold. It uses the double thresholds algorithm to filter maxima in the high-frequencysub-band, and obtains two matrices to compensate the link to reduce the pseudo-edge. As thetransform coefficients of non-sampling Contourlet the structural characteristics, there is richedge information in the low frequency sub-band, uses Canny operator to detect the lowfrequency sub-band, effectively controls the noise, and eliminates the pseudo-edge image for image understanding and segmentation.Thirdly, this article points out the image segmentation algorithms, based on Contourletdomain Hidden Markov Tree Model in accordance with the context structure.Wavelet-domain Hidden Markov Tree Model Image segmentation results will be easier toexist the direction of the edge components prone to ambiguity and singularity of diffusion,Contourlet transform can fully capture the higher dimensional and singular image, to obtainthe initial scale the beginning of segmentation by Contourlet domain hidden Markov treemodel, to use the context of the scale adaptive fusion method, to integrate into the smallestscale, that is, pixel-level segmentation from the appropriate coarse scale segmentation results,in order to obtain the final result of image segmentation and the desired results. Lay a solidfoundation for further recognition. The goal of segment the information can also betransimitted rapidly, and the remote real-time monitor can be realized.Finally, this article proposes reconstruction method, based on Contourlet transform, oforthogonal matching pursuit compressed sensing image. Compressed sensing system uses apriori knowledge by sparse representation of the image, to restructure the original image froma small number of observations, thus to break the limit of sample rate of the Nyquist samplingtheorem. Currently the compressed sensing systems usually use orthogonal wavelet with threedirections to get image, applies the iterative shrinkage method for solving the problem ofcorresponding optimization. The shortage of this method is slow convergence speed, and thereconstruction image has obvious pseudo-Gibbs effect. Coefficients of Contourlet transformis sparser than the wavelet coefficients, and can reconstruct with fewer coefficients of thesame quality images, and through orthogonal matching pursuit force iterative termination inorder to increase the efficiency. It uses complex calculations to make up for short-board ofhardware resources to reduce the requirements of accessing panoramic images of thehardware resource.