Revitalizing image retrieval: AI enhancement and metaheuristic algorithm adaptation

Abstract

INTRODUCTION: In recent years, development of digital technology has led to number of images, which can be stored in digital format. However, searching and retrieval of images in large image DB (Database) is a mammoth task. Therefore, different image retrieval techniques have been used for retrieving the suitable images, which includes retrieval of images using keywords or annotations, however, these methods are considered to be time consuming and leads to imprecise outcome.

OBJECTIVES: Therefore Effective and precise retrieval of suitable images from huge DB can thrived by utilizing CBIR (Content Based Image Retrieval) system. However, incorporation of CBIR in most existing studies resulted in low accuracy for IR. So, proposed model incorporates Modified ResNet50 (M-ResNet50) and VGG 16 model for feature extraction in order to extract the best features as M-ResNet50 utilizes extra dense layers which aids in better feature extraction process.

METHODS: After feature extraction, the features are fused using PCA and fed to Modified PSO (M-PSO) model for obtaining optimized features since M-PSO is fast and aids in selecting optimal features after processing from insignificant features that primarily set the preferred number of necessary features.

RESULTS: Moreover, M-PSO require less parameters to tune instead of a huge number of parameters by incorporating K parameters of KNN algorithm in order to find the nearest images to Query Images (QI), thereby making the model appropriate for IR process with better similarity score.

CONCLUSION: The proposed model utilizes 8 different sun images at different intervals for IR process. Finally, the proposed model is evaluated by using several metrics such as accuracy, precision, recall and F1 score, besides the proposed model is compared with various existing models in order to evaluate the efficiency of the proposed model.