Deep Joint Source-Channel Coding for Person Re-Identification With Related Image Sources

In this paper, we propose a deep learning based joint source-channel coding (JSCC) scheme for wireless transmission of multiple related image sources in person re-identification tasks. Our network utilizes parameter sharing for the encoder at the transmitter and adopts a joint decoding architecture for the decoder at the receiver, reducing storage space and computational resource consumption. By leveraging object-related information from multiple perspectives, the decoder fuses richer features to perform corresponding re-identification tasks. Experimental results demonstrate that the performance of our proposed method degrades gracefully as available bandwidth resources decrease, and the proposed multi-transmitter JSCC scheme achieves significant gains under low bandwidth and low signal-to-noise ratio conditions, compared with the existing approach. This work shows the potential of the JSCC-based approach for efficient and robust wireless transmission of related image sources in person re-identification scenarios.