Fig. 6. Implementation of edge detection.

a Schematic illustration of the operations for the edge detection. The convolution is implemented by sequentially presenting the sub-images whose pixel values are translated to the illumination signals to the kernels based on FE-PS-NET. The merging, normalization, and binarization of output currents are performed in software. b From left to right: theoretical dimensionless weights, theoretical photoresonsivities scaled from dimensionless weights, actual weights right after programming, and actual photoresonsivities after edge detection, for Kernel 1 and 2. The unit of photoresponsivity is mA/W (not shown). c Output currents I₁ (left panel) and I₂ (right panel) after the convolution with the two kernels. The values outside the brackets are the actual I₁ and I₂ values while those in the brackets are the differences between the theoretical I₁ and I₂ values and their corresponding actual values. d Image obtained by merging I₁ and I₂ (left panel) and final output image after normalization and binarization showing the detected edge (right panel).