Parallel memory systems for image processing

Kichul Kim, V. K. Prasanna Kumar

Abstract: A novel memory system is proposed for image processing. Latin squares, which are well known combinatorial objects, are used as the skew function of the memory system. A Latin square with desired properties for image-array access is introduced. The resulting memory system provides access to various subsets of image data (rows, columns, diagonals, main subsquares, etc.) without memory conflict. The memory modules are fully utilized for most frequently used subsets of image data. The address generation can be performed in constant time. This memory system achieves constant-time access to rows, columns, diagonals, and subarrays using a minimum number of memory modules.<