Compression And Compressed Sensing Reconstruction Based On Contourlet Transform For Medical Images

Medical imaging equipments generate a large number of medical images. Due to the characteristics of multi-quantization levels, high-resolution and high-quality, there are enormous challenges for medical images' transfer and storage. Therefore, medical images compression is one of the important and urgent problems to be solved.The lossless compression is often used in that the diagnostic details need to be preserved. Although it's possible to restore the original image from the compressed data accurately by this method, the compression efficiency is too low, to meet the demand of medical images compression. In this thesis, a hard threshold algorithm is proposed to compress medical images based on contourlet transform, with the performance compared to wavelet transform. Experimental results demonstrate that the peak signal to noise ratio(PSNR) based on contourlet transform is higher and the image quality is better. However, This algorithm can't highlight the details of region of interest(ROI). An algorithm is proposed for ROI compression for medical images based on contourlet transform and set partitioning in hierarchical trees(SPIHT), and experimental results demonstrate its effectiveness through the qualitative and quantitative analyse. In addition, the curvelet transform and SPIHT method are combined for ROI compression of medical images to veryfy the effects of multiscale geometric analysis(MGA). The compression performance of the curvelet transform is also effective. However, the curvelet transform is proposed in the continuous domain originally, then applied to the discrete image gradually, and it's too complicate to implementation. On the other hand, the contourlet transform is defined in the discrete domain directly, and it's more suitable for two-dimensional discrete images. Therefore, the contourlet transform is used as sparse representation method in this thesis.Compressed sensing(CS) is a novel theoretical framework for information acquisition and processing, which is different from traditional compression methods in essence. Taking advantage of the sparsity inherent in real world signals, CS can be used to reconstruct the original image accurately from non-adaptive and small linear measurements by nonlinear optimization methods. Therefore, the advantages of contourlet transform and CS are combined and proposed in this thesis. The nonsubsampled contourlet transform(NSCT) is applied to sparse represent ation of the original images, then Fourier matrix is used as the measurement, finally the iterative soft threshold algorithm(ISTA) is used to reconstruct the medical images. Experimental results demonstrate that this method can reduce the sampling rate effectively and reconstruct the medical images with high quality, which can be extended and widely used in rapid medical imaging technology.