This paper presents a novel in situ image cryptography scheme for neuromorphic vision sensors based on all-optically controlled (AOC) memristors. The AOC memristors exhibit bidirectional persistent photoconductivity, allowing visual images to be stored, encrypted, decrypted, denoised, and destroyed within the sensor. The encrypted images are secure even with trained neural networks, and the decrypted images can be encoded and recognized accurately using a memristive neural network. The proposed scheme integrates multiple functions, including image sensing, cryptographic computing, and processing, into a single sensor, providing a simple and efficient solution to the security challenges faced by vision sensors. The effectiveness of the scheme is demonstrated through experiments, showing high recognition accuracy and robustness against hacking attacks. This approach offers a promising solution for securing visual information in lightweight edge devices with limited computational power.This paper presents a novel in situ image cryptography scheme for neuromorphic vision sensors based on all-optically controlled (AOC) memristors. The AOC memristors exhibit bidirectional persistent photoconductivity, allowing visual images to be stored, encrypted, decrypted, denoised, and destroyed within the sensor. The encrypted images are secure even with trained neural networks, and the decrypted images can be encoded and recognized accurately using a memristive neural network. The proposed scheme integrates multiple functions, including image sensing, cryptographic computing, and processing, into a single sensor, providing a simple and efficient solution to the security challenges faced by vision sensors. The effectiveness of the scheme is demonstrated through experiments, showing high recognition accuracy and robustness against hacking attacks. This approach offers a promising solution for securing visual information in lightweight edge devices with limited computational power.