ASPL-APBT Algorithm for Block Compressed Sensing Image Reconstruction

Aiping Yang, Gai Li, Zhengxing Hou, Yuqing He
2014 Journal of Communications  
Based on the block compressed sensing (BCS) framework, a new and non-orthogonal transform named all phase biorthogonal transform (APBT) is introduced to exploit the image sparsity, reduce the encoding complexity and be applicable to the blocked image easily. APBT exploits the signal sparsity better than DCT, and meanwhile it overcomes the defects of multiscale transform such as wavelet transform with high computational complexity and the feature of not being applicable to the blocked image. In
more » ... rder to improve the efficiency of BCS reconstruction, the accelerated smoothed projected Landweber (ASPL) iteration algorithm is put forward. Combined with the sparse constraints in APBT, the BCS-ASPL-APBT reconstruction algorithm is advanced. Experimental results demonstrate that the proposed algorithm outperforms the method of using DCT sparsifying coupled with common SPL iteration not only in the aspect of PSNR, but in terms of the reconstruction time and the iteration number. Index Termsaccelerated SPL I. INTRODUCTION Compressed sensing (CS) has gained increasing interests over the past few years. The CS theory demonstrates that signals with sparse representation under some transform domain can be precisely reconstructed only from a small set of measurements [1]-[3]. As applied to 2D images, however, CS faces several challenges including a computationally expensive reconstruction process and huge memory required to store the random sampling operator. Recently, several fast algorithms [4], [5] have been developed for CS reconstruction, but with the reconstruction quality as the sacrifice. With the purpose of resolving huge memory for sensing matrix, the structurally random sensing matrixes were addressed in [6], [7], but these matrixes have poor universality. So the block compressed sensing (BCS) theory was advanced in [8], [9], in which the block-based sampling operation was used and the smoothed projection-based Landweber iterations (SPL) was proposed to accomplish CS reconstruction aiming at improving the reconstructed-image quality by eliminating blocking artifacts. In recent years SPL based methods Manuscript
doi:10.12720/jcm.9.4.371-378 fatcat:7efdmoqewrf63ascbrnpiyhypy