Numerical Neuromorphic Vision and Optical Character Recognition without Floating-Point or Convolutional Networks

Computer vision has become synonymous with deep Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), requiring high-precision floating-point operations and millions of parameters. In this paper, we present an alternative vision paradigm designed entirely around integer-only neuromorphic operations. By mapping input pixels directly to localized 2D integer grids (active sensory boards), the system performs edge detection, shape orientation, and optical character recognition (OCR) using simple spatial voting patterns. Classification is achieved by intersecting active pixel coordinates with stored character sketches using binary voting. The entire pipeline operates with 8-bit integers (uint8_t) and additions, avoiding all matrix multiplications, backpropagation, and floating-point parameters.