研究目的
To propose a novel Deep Linear Discriminant Analysis Hashing(DLDAH) algorithm for efficient and scalable image retrieval by minimizing semantic loss during the hashing process.
研究成果
The proposed DLDAH algorithm shows superior performance in image retrieval by efficiently mapping images to compact binary codes, thus enabling fast retrieval. The method is scalable and adaptable to different classification grains.
研究不足
The algorithm's performance is dependent on the quality of feature extraction and the size of the dataset. The time consumption during the offline phase is high due to the use of deep networks.
1：Experimental Design and Method Selection:
The study involves two main stages - Hash label generation and Deep hash model construction. The first stage uses Linear Discriminant Analysis(LDA) to map image features into hash labels. The second stage trains a simple deep learning network using these hash labels for image hashing.
2：Sample Selection and Data Sources:
The experiments are conducted on benchmark datasets CIFAR-10, CIFAR-100, and STL-
3：List of Experimental Equipment and Materials:
AlexNet and GoogLeNet are used for feature extraction. MLP and Slice are used for constructing hash function models with Sigmoid and BatchNorm as activation functions.
4：Experimental Procedures and Operational Workflow:
Features are extracted using pre-trained models, hash labels are generated, and a deep hash model is trained. The model is then used for image retrieval.
5：Data Analysis Methods:
Precision and recall metrics are used to evaluate the performance of the algorithm.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
