Abstract: We present in this paper a hardware implemented system which allows compression and decompression of 256 $MUL 256 still-images at almost video rate. We have followed a scheme inspired from those presented by Barland. In the compression stage, the main steps are: biorthogonal wavelet transform (5 levels), scalar and vector quantization of the last two levels, lossless coding; the decompression stage follows the same steps in their inverse forms. We have tested the robustness of the reconstruction in the presence of truncatures made on the wavelet coefficients, and it seems that the use of wavelet with high regularity filter is not really important in the compression stage. Finally we have chosen FIR splines filters with 3 coefficients for analysis and 5 for synthesis. The wavelet coefficients are truncated and coded in a non linear way. The codebook is created once for all with a set of training images using an algorithm inspired from Chok-Ki Chan. The hardware implementation has required 4 FPGA circuits and some RAM. For a 256 $MUL 256 image with gray levels coded on bytes the compression time is 64 ms while the decompression time is 17 ms with a compression ratio of 87 and a mean PSNR of 28 dB.!9
展开▼
机译：摘要：我们在本文中展示了一个硬件所实施的系统，它允许压缩和减压在几乎视频速率下静止图像256 $ MUL 256静止图像。我们遵循了一个受欢迎的计划，这些计划受到巴兰呈现的那些。在压缩阶段，主要步骤是：双正交小波变换（5级），标量和矢量量化的最后两个级别，无损编码;减压阶段以其反向形式遵循相同的步骤。我们在在小波系数上制作的触头存在下测试了重建的稳健性，并且似乎使用具有高规律性滤波器的小波在压缩阶段中非常重要。最后，我们选择了具有3系数的FIR拼接过滤器，用于分析和合成5。小波系数以非线性方式截断并编码。编码本，为所有人创建一次，其中包含一组训练图像，该算法激发了来自Chok-Ki Chan的算法。硬件实现需要4个FPGA电路和一些RAM。对于256 $ MUL 256图像，灰色级别编码在字节上，压缩时间为64 ms，而减压时间为17毫秒，压缩比为87，平均psnr为28 dB。！9
展开▼
