Exponential Sailfish Optimizer-based generative adversarial network for image annotation on natural scene images

Abstract

Generally, automatic image annotation can offer semantic graphics for recognizing image contents, and it creates a base for devising various techniques, which can search images in a huge dataset. Although most existing techniques mainly focus on resolving annotation issues through sculpting tag semantic information and visual image content, it ignores additional information, like picture positions and descriptions. The established Exponential Sailfish Optimizer-based Generative Adversarial Networks are therefore used to provide an efficient approach for image annotation (ESFO-based GAN). By combining Exponentially Weighted Moving Average (EWMA) and Sailfish Optimizer (SFO), the ESFO is a newly created design that is used to train the GAN classifier. Additionally, the Grabcut is presented to successfully do image annotation by extracting the background and foreground images. Additionally, DeepJoint segmentation is used to divide apart the images based on the background image that was extracted. Finally, image annotation is successfully accomplished with the aid of GAN. As a result, image annotation uses the produced ESFO-based GAN's subsequent results. The developed approach exhibited enhanced outcomes with maximum F-Measure of 98.37%, maximum precision of 97.02%, and maximal recall of 96.64%, respectively, using the flicker dataset.