基于SPIHT图象压缩编码的Turbo codes编码

Turbo-codes Used For Progressive Image Based On SPIHT

针对信源与信道编码的特点 ,提出了结合SPIHT(setpartitioninginhierarchicaltrees)算法和改进的TurboCodes的信源与信道联合编码方案 .这一方案充分考虑了SPIHT算法和TurboCodes的特性 ,根据信源编码后数据在重建时的重要程度 ,进行不同等纠错保护的信道编码 .实验表明 ,论文的方法能够使误码率与码长达到一个较好的平衡 。

Source and channel coding have been treated separately in most cases. But it can be observed from source coding of images that each part of compressed images data has a different sensitivity to error. In this paper, SPIHT (Set partitioning in Hierarchical trees) is adopted for source coding. The advantage of the method is that the bits in the compressed data are generated in the order of importance, and the important bits which have more information for reconstruction are at the front of the bit stream. And channel coding adopts Turbo Codes which is a newly developed channel coding technique with amazing error correcting capability. This paper presents a joint source and channel coding scheme based on SPIHT and revises Turbo codes according to the characteristic of source and channel coding. It takes full advantage of the characteristics of SPIHT and Turbo Codes. And channel coding adopts unequal error protection (UEP) for the compressed data according to the importance for reconstruction. For an image, it may be advantageous to use a strong error correcting code to protect its important information and a weaker code for the other data. The experiment uses Lena256×256 and rocket256×256 to compare the performances between equal error protection (EEP) and UEP. The two images are coded by EEP and UEP after being source coded. For the EEP code, the channel code rate is 1/2 for both header and image data. For the UEP code, the channel code rates are 1/3 and 1/2 for the important information including header syntax and image important data and the other image data respectively. In addition, different times of iterations were performed for the decoding scheme of UEP (8 iterations for the important information and 5 iterations for the other data). UEP gives better reconstructive images than EEP when SNR is 1.0 dB and 0.7 dB. The experimental results show that the proposed scheme could make a better balance between bit error rate and code length. It is found that Turbo Codes combined with unequal error protection give excellent performance for compressed image transmission, even at lower SNR.