Aiming at two major drawbacks existed in the single image non-uniformity correction based on compressive sensing, which is that fairly rich information of high frequency sub bands could not be decomposed by wavelet transform and the sparse degree of images need to be known through (ROMP). A based on wavelet packet transform and sparsity adaptive compression sampling matching pursuit (CoSaSAMP) algorithm was proposed to reconstructed the image and corrected the infrared images. The improved method sparsed image by wavelet packet transform, corrected the extracted 25% pixels from original infrared image by point-sampling matrix through the improved midway infrared e-qualization algorithm. The missing pixels was reconstructed by CoSaSAMP algorithm. The experimental results show that the proposed method was further improved in root-mean-square error ( RMSE) and peak signal to noise ration ( PSNR) compared with the single image non-uni-formity based on wavelet transform, the RMSE is reduced to nearly 30%, the average between each line closes to ideal image, image recon-struction are of good quality.%针对单幅图非均匀校正中小波变换不能分解红外图像中相当丰富的高频子带以及正则化正交匹配追踪(ROMP)重构算法需要已知红外图像的稀疏度等问题,提出了一种基于小波包变换结合稀疏度自适应压缩采样匹配追踪算法(CoSaSAMP)实现对图像的重构,从而达到校正图像的目的.该方法利用小波包变换对原图像稀疏,将点样本矩阵作为测量矩阵,提取原红外图像的25%数据,利用改进的中值直方图算法校正提取的数据,然后利用CoSaSAMP重构图像.研究结果表明:与基于小波变换压缩感知的非均匀校正相比,本算法在均方根误差、峰值信噪比方面都得到了进一步的改善,均方根误差降低了30%左右,列间均值更加接近理想校正图像效果,图像重构质量好.
展开▼
