Adaptable Area-Efficient Parallel Architecture for Grey and Color Image Convolvers

Window based spatial domain two dimensional (2-D) image convolution is a widely used algorithm in many image processing applications including intrusion detection. Systolic array architectures have been widely used to implement 2-D convolvers so as to exploit the hardware parallelism on dedicated hardware platforms and thereby increase the throughput of the system. Higher levels of parallelism have not been tried for 2-D convolvers due to obvious increase in area. We propose an adaptable area-efficient parallel architecture which can be used to increase the throughput. In this paper, we study the advantages of the novel area-time efficient look-up table (LUT) based method for constant coefficient multiplication proposed in our earlier work with increasing levels of parallelism. The area reduction with this architecture increases to about 36% from 1% when the level of parallelism is increased to 8 from 1 on 0.18 um technology. Further, we also evaluate the advantages of the proposed parallel architecture.