Fast Implementation of Two-Dimensional Digital Filter using Block Processing Based on Separate Computation of Zero-State and Zero-Input Responses

提出了快速二维数字滤波器的多处理器实现方法和有效的滤波算法。图像平面被划分为P*Q个样本的块，并且同时执行块内和块间并行处理。块内处理基于Roesser的状态空间模型。块间并行处理的同步。考虑双向（水平和垂直）和单向（垂直）环耦合处理系统。从一个块计算的水平和垂直状态变量分别被传送到下一个右侧和顶部块。建立了各处理系统无等待接收状态变量的定时条件.块内处理算法：为了提高处理速度，减少每个样本的运算次数，提高并行计算的程度，提出了零状态和准零状态块处理方法，并与块处理方法进行了比较。曲塞块体法优于其他方法，具有上级的特点.分母可分传递函数的高效处理系统。由于滤波器的块状态矩阵具有高稀疏性，因此将行和列分别互换，以减少矩阵大小。结果，系统的吞吐量有所提高，所需处理器的数量有所减少。单向环耦合系统的优化设计单向系统比双向系统具有更低的帧延迟，其中帧延迟定义为从一帧处理开始到结束的时间。用最少的处理机数实现指定吞吐量的系统最佳设计条件，即：例如，块的形状和大小、同时处理的块的数目通过计算机模拟来找到。

Multi-processor implementation of fast two-dimensional digital filters and efficient filtering algorithms are developed. Picture plane is partitioned into blocks of P*Q samples and intra- and inter-block parallel processing s are simultaneously performed. The intra-block processing is based on Roesser's state-space model.1. Synchronization of inter-block parallel processing.Bi-directional (horizontal and vertical) and one directional (vertical) ring coupled processing systems are considered. The horizontal and vertical state variables computed from one block are transferred to the next right and top blocks respectively. Timing condition that each processing system has no waiting to receive state variables is established.2. Algorithms for intra-block processing.To increase the processing speed with decreasing of the number of operations per sample and with increasing parallel computation degree, zero-state and quasi zero-state block processing methods are proposed and compared with block processing method. Qusai block method is found to be superior to others.3. Efficient processing system for denominator separable transfer function.As the block state matrix of the filter has high sparsity, the rows and columns are interchanged respectively to reduce the matrix size. as a result, some raising of the throughput of the system and some reduction of the number of required processors are achieved.4. Optimum design of one directional ring coupled system.One directional system has advantage of lower frame latency than bi-directional system, where frame latency is defined as the time from beginning to the end of one frame processing. Optimum design conditions of the system to realize assigned throughput using minimum number of processors , i. e., the form and size of a block, number of simultaneously processed blocks, are found by computer simulation.