多标记收缩哈希方法的人脸属性检索*

哈希方法具有存储成本低、查询速度快等优点.但是,大多数哈希方法只用于处理单标记问题,较少用于图像多标记问题.因此,文中提出多标记收缩哈希方法(MLCH),可以保留多标记图像的多水平语义相似性.MLCH利用数据的多标记作为监督信息,依据优化选择算法选取三元组,并加入收缩约束项以获取更饱和的二进制编码.在数据库CelebA和PubFig上评估MLCH性能,证明收缩约束项的有效性,并证明MLCH在大规模图像检索任务中的优越性.

Multi-label Contractive Hashing Method for Face Attributes Retrieval

Hashing methods possess advantages of low storage cost and fast query speed. However, most of the current hashing methods are designed to handle simple binary similarity rather than the complex multilevel semantic structure of the images associated with multiple labels. In this paper, a multi-label contractive hashing method(MLCH) is proposed to preserve the multilevel semantic similarity of multi-label images. In particular, the supervising information of attributes is proposed to help the training of the model and adopt an optimized selection algorithm to select training samples. Meanwhile, a contractive constrain term is added to the loss function to improve the quality of the generated binary codes. The proposed approach is evaluated on CelebA and PubFig databases, and the experimental results demonstrate its superiority over several state-of-the-art hashing methods on the task of large-scale image retrieval.