中文摘要：

提出了一种图像小面模型拟合的快速并行计算结构。推导了以离散切比雪夫正交多项式为基底的图像小面模型拟合过程，分析了其计算结构和内在的并行性。计算结构包括滑动窗产生单元、多路卷积单元和拟合单元，整体为流水线结构，而后两者均采用空间并行结构。基于小面模型的序列图像插值实验结果表明，该计算结构占用存储空间少、数据延迟小、实时性强，提高了小面模型拟合的实用性。

英文摘要：

A parallel computing architecture for image facet model fitting is presented. The image facet fitting procedure is deduced by using discrete Chebyshev orthogonal polynomial. The computing architecture and the inherent parallelism are analyzed. The architecture comprises the moving window generator, the parallel convolver and the function fitting unit. The whole architecture is arranged by the pipelined mode, while the last two steps by the data parallelism mode. Facet-based image interpolation experimental results show that the proposed computing architecture has little memory usage, low data latency, and real-time performance, thus improving the practicability of facet model fitting.