自适应多尺度分块压缩感知算法

发布时间：2018-07-31 14:10

【摘要】：目的基于小波域的多尺度分块压缩感知重构算法忽略了高频信号在重构过程中的作用,丢失了大量的边缘与细节信息。针对上述问题,提出一种自适应多尺度分块压缩感知算法,不仅合理利用低频信息还充分利用图像的高频信息,在图像细节复杂度提高的情况下保证图像重构质量的提高。方法首先进行3层小波变换,得到一个低频信号和9个高频信号,分别进行小波逆变换后分成大小相同互不重叠的块,对低频部分采用2维邻块边缘自适应加权滤波的方法进行处理,对高频部分采用纹理自适应分块采样,最后利用平滑投影Landweber(SPL)算法对其进行重构。结果与已有的分块压缩感知算法、基于边缘和方向的分块压缩感知算法和基于纹理和方向的分块压缩感知算法相比,本文算法在不同的采样率下,性能均有所提升,代表细节信息的高频信号得到充分重建,改进的算法所得到的重建图像具有较高的分辨率,尤其对细节较为丰富的图像进行重建后具有较高的峰值信噪比;2维邻块边缘自适应加权滤波有效的去除了重建图像的块效应,且重建时间平均减少了0.3 s。结论将三层小波变换后的高频分量作为纹理部分,利用自适应多尺度分块重建出图像的轮廓与边缘;将低频分量直接视为平坦部分,邻块边缘自适应加权滤波重建出图像细节,不仅充分利用了图像的高低频信息,还减少了平坦块检测过程,使得重建时间有效缩短。经实验验证,本文算法重建图像质量较好,尤其是对复杂图像明显消除了块效应,边缘和纹理细节较清晰。因此主要适用于纹理细节较复杂的人脸图像、建筑图像和遥感图像等。
[Abstract]:Aim Multi-scale block compression perceptual reconstruction algorithm based on wavelet domain neglects the function of high frequency signal in the reconstruction process and loses a lot of edge and detail information. To solve the above problems, an adaptive multi-scale block compression sensing algorithm is proposed, which not only makes rational use of the low frequency information but also makes full use of the high frequency information of the image, so as to ensure the improvement of the image reconstruction quality with the increase of the complexity of the image details. Methods three layers of wavelet transform were carried out, one low frequency signal and nine high frequency signals were obtained. After inverse wavelet transform, they were divided into blocks with the same size and no overlap. The low-frequency part is processed by 2-D adjacent block edge adaptive weighted filtering, the high-frequency part is sampled by texture adaptive block sampling, and the smooth projection Landweber (SPL) algorithm is used to reconstruct it. Results compared with the existing block compression sensing algorithm, the edge and direction-based compression sensing algorithm and the texture and direction-based block compression sensing algorithm, the performance of this algorithm is improved at different sampling rates. The high frequency signal representing the detail information is fully reconstructed, and the reconstructed image obtained by the improved algorithm has higher resolution. Especially for the image with rich details, it has a high peak SNR and 2-dimensional edge adaptive weighted filtering to remove the block effect effectively, and the reconstruction time is reduced by 0.3 s on average. Conclusion the high-frequency component of three-layer wavelet transform is taken as texture part, and the contour and edge of image are reconstructed by adaptive multi-scale block, and the low-frequency component is regarded as a flat part directly, and the image details are reconstructed by adaptive weighted filter of adjacent block edge. It not only makes full use of the high and low frequency information of the image, but also reduces the process of flat block detection, which effectively shortens the reconstruction time. Experimental results show that the proposed algorithm has better image quality, especially for complex images, the block effect is eliminated, and the edge and texture details are clear. So it is mainly applied to face image, building image and remote sensing image with complicated texture details.
【作者单位】：中国矿业大学信息与控制工程学院;
【基金】：国家自然科学基金项目(U1261105)~~
【分类号】：TN911.73

【相似文献】