An Architectural Study of a Massively Parallel Processor for Convolution-Type Operations in Complex Vision Tasks

Complex vision tasks, e.g., face recognition or wavelet based image compression, impose severe demands on computational resources to meet the real-time requirements of the applications. Clearly, the bottleneck in computation can be identified in the first processing steps, where basic features are computed from full size images, as motion cues and Gabor or wavelet transform coefficients. This paper presents an architectural study of a vision processor, which was particulary designed to overcome this bottleneck.