Single-shot incoherent multifunctional image differentiation with spatial tiling optical convolution

Incoherent optical spatial differentiation may allow high speed, low computational power and low latency image processing. However, it is a great challenge to directly design an optical convolutional kernel containing negative values for image differentiation. Therefore, polarization or wavelength-multiplexed bipolar point spread functions (PSFs) followed by digital subtraction is proposed to achieve incoherent optical edge detection. Unfortunately, these methods will increase the device design complexity and only support single isotropic edge detection. Here we proposed and experimentally demonstrated a single-shot incoherent multifunctional image differentiation with spatial tiling optical convolution. As a proof of concept, we specially designed four spatial tiling non-negative optical convolutional kernels based on 4f system through inverse design method, and simultaneously realized three types of spatial differential operators. The x and y directions spatial differentiation and isotropic edge detection can be experimentally demonstrated simultaneously after digital subtraction. Our solution is capable of relaxing device complexity, increasing scalability and facilitating intelligent design, which is of great significance for accelerating machine vision tasks under incoherent illumination.