An Analysis of Two Novel and Efficient Deep Learning Models for Fast and Accurate Image Retrieval

Due to the presence of various limitations in traditional machine learning algorithms, the interest of the researcher community has been shifted towards Deep learning. In this paper, two Deep learning methods namely Deep belief network (DBN) and Stacked autoencoder (SAE) have been analyzed for image retrieval task. But, in order to retrieve images from vast storehouses, more retrieval time is utilized. To solve this issue of retrieval time, two different indexing techniques namely Similarity-based indexing (SBI) and Cluster-based indexing (CBI) have been used. Thus, four models namely DBN-SBI, DBN-CBI, SAE-SBI and SAE-CBI have been developed and tested on two benchmark datasets, which are MIT-Vistex and INRIA-Holidays. Among these models, DBN-SBI obtains the highest results in terms of Precision, Recall and Retrieval time. Average precision of 98.45% and 86.53% with retrieval time of 0.035 seconds and 0.149 seconds has been obtained on Vistex and Holidays dataset respectively which is higher than many state-of-the-art related models.

Automated summary

Research has shown that deep learning methods are more effective for image retrieval tasks compared to traditional machine learning algorithms. By comparing the performance of four different models on two benchmark datasets, the results show that the DBN-SBI model performs best in terms of precision, recall, and retrieval time.