北京理工大学电子工程系,北京,100081
纸质出版:2003
移动端阅览
赵保军, 史彩成, 毕 莉, 等. 基于FPGA和DSP实现的实时图像压缩[J]. 电子学报, 2003,31(9):1317-1319.
ZHAO Bao-jun, SHI Cai-cheng, BI li, et al. Implementation of Real-Time 2D-DCT with FPGA and DSP[J]. Acta Electronica Sinica, 2003, 31(9): 1317-1319.
利用FPGA的并行分布流水特点
选用exilinx公司的50万门级芯片XCV400E
设计并实现CIF格式(352×288象素)图像实时DCT变换.该设计采用乒乓模式
只需设计一个快速算法模块
(F×C
T
)就解决了C×F×C
T
的实现算法.当视频信号通过数字化后逐行输入FPGA
在行、场同步信号和采样时钟的控制下
每输入一组数据(8个)
就进行行向量与
C
T
的矩阵乘运算(
F×C
T
)
并将结果按转置方式保存
每输入一个数进行一次(1×8)×(1×8)矩阵运算
每行进行352×(1×8)×(8×8)次矩阵运算
其中44次(1×8)×(8×8)矩阵运算的结果需要按转置形式(
H
T
=(F×C
T
)
T
)存储;当输入下一组8行数据时
对该组数据进行与前述8行数据相同的矩阵运算
而对刚做完(
F×C
T
)运算的8行相应结果
则按正常顺序取出进行(
H
T
×C
T
)运算
将结果按转置形式(
G
T
=C×H
)输出.从而以实时流水的方式完成
C×F×C
T
运算.功能仿真、时序仿真和与TMS320C62X系统的成功对接验证了本设计及算法的正确性.
Because of the FPGA's parallel pipelining processing features
this paper designed and implemented the real time CIF format image DCT using Exilinx Company's 500000 gate grade chip XCV400E.Using ping-pong model
C×F×C
T
is implemented only by designing one fast algorithm model (
F×C
T
).Digital video signal is input to FPGA line by line.Controlling by horizontal sync and vertical sync
every group data of 8 pixels as a vector is inp
ut and is multiplied
C
T
i.e.(
F×C
T
).The computed results are stored as transform format.Each pixel needs one (1×8)×(8×8) matrix operations.Each line needs 352×(1×8)×(8×8) times matrix operations.44times (1×8)×(8×8) matrix operations results need storing as transform format (
H
T
=(F×C
T
)
T
).When next 8 line data are input
they are processed with the same way as the above.For the last 8 line's first processed results (
F×C
T
)
they are read out and processed as (
H
T
×C
T
).The final results are output as transform format (
G
T
=C×H
).Therefore
the continuous real-time whole field pix DCT transform
C×F×C
T
is finished.Function and timing simulation and the successful connection with TMS320C62X system verified the design and implement.
0
浏览量
1806
下载量
11
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621