Evolutionary Optimization with Deep Transfer Learning for Content based Image Retrieval in Cloud Environment

Recently, image model in cloud computing (CC) grabbed great deal of interest to guarantee confidentiality of data and secure data communication among clients cloud storage and cloud server, and end-users. Conventional secure image retrieval (IR) approaches are not suitable to adopt a large-scale IR in cloud environment. In the field of computer vision (CV), an effectual representation of image feature vectors for image retrieval will remain a significant issue. various research was conducted on CBIR by utilizing machine learning (ML) techniques and numerous descriptors. This article designs a Dwarf Mongoose Optimization with Transfer Learning for Content based Image Retrieval in Cloud Environment, named DMOTL-CBIRC. The presented DMOTL-CBIRC technique intends to retrieve similar images in the cloud sever based on QIs. It follows a three stage process namely feature extraction, parameter tuning, and similarity measurement. Initially, the presented DMOTL-CBIRC technique exploits SqueezeNet method for an effectual generation of feature vectors (FVs) for the input image and QIs. Besides, the DMO technique is used for the hyperparameter tuning of the SqueezeNet model. For similarity measurement, Manhattan distance is applied which retrieves the highly similar images. The experimental evaluation of the DMOTL-CBIRC technique is investigated on Corel10K dataset and the outcomes highlighted the improvements of the DMOTL-CBIRC technique over other models.