Multi-proxy deep hashing for image retrieval

Deep hashing is an effective method for content-based image retrieval due to its low storage requirements and fast retrieval speed. However, deep hashing is known to suffer from semantic information loss and quantization errors due to the need to transform high-dimensional feature vectors into low-dimensional binary codes, which will inevitably affect the retrieval performance. To address this problem, we propose a novel multi-proxy deep hashing (MPDH) method, which is able to unify two high-quality hash codes into a single deep model. Meanwhile, we present a joint retrieval strategy that amalgamates the contributions of each hash code to enhance retrieval performance. Furthermore, we introduce a novel proxy quantization strategy to reduce the inconsistency between continuous feature vectors and binary hash codes. We evaluate our approach on three public benchmark datasets, including ImageNet, NUM-SIDE, and MS-COCO. Experimental results demonstrate the effectiveness of our approach, achieving state-of-the-art performance on all benchmark datasets.